This thesis has been submitted in fulfilment of the requirements for a

postgraduate degree (e. g. PhD, MPhil, DClinPsychol) at the University of

Edinburgh. Please note the following terms and conditions of use:

• This work is protected by copyright and other intellectual property rights,

which are retained by the thesis author, unless otherwise stated.
• A copy can be downloaded for personal non-commercial research or
study, without prior permission or charge.
• This thesis cannot be reproduced or quoted extensively from without
first obtaining permission in writing from the author.
• The content must not be changed in any way or sold commercially in
any format or medium without the formal permission of the author.
• When referring to this work, full bibliographic details including the
author, title, awarding institution and date of the thesis must be given.
Creativity and Semantic Control: Cognitive and Neural Insights
from Divergent Thinking to Discourse Production

Tanvi Patel

2024

A thesis submitted to the University of Edinburgh for the degree of Doctor of Philosophy in
Psychology

1
 For my Grandfather,

who was so pleased that I was carrying on the torch .

2
 Signed Declaration

I, Tanvi Patel, declare:

(a) that the thesis has been composed by solely by myself;

(b) the work has been solely my own;
(c) this work has not been submitted, in whole or in part, for any other degree or
professional qualification;

Parts of this work have been submitted, or are in the process of being submitted, for
publication.

Patel, T., MacPherson, S. E., & Hoffman, P. (2024). Investigating age related differences in
semantic control mechanisms involved in creative cognition (Submitted to Memory and
Cognition)

Patel, T., MacPherson, S. E., & Hoffman, P. (2024). Dissecting the RAT: Exploring the
dynamics of semantic search in creative thinking. (In Preparation, Journal of Creative
Behaviour)

3
 Abstract

Effective semantic processing is governed by two interacting neural systems: the

representational system and the semantic control system. The representational system stores
semantic knowledge, while the semantic control system manipulates this stored knowledge to
extract task-relevant information. These flexible retrieval mechanisms are essential for various
cognitive functions, including creativity. Creativity is believed to stem from a dynamic
interplay between bottom-up associative processes, which rely on semantic memory during
idea generation, and top-down executive control processes during idea evaluation, ensuring
goal maintenance and inhibition of dominant responses. Despite neuroimaging studies
underscoring the importance of semantic control regions in creativity, this area remains
underexplored in the cognitive literature.
This thesis explores the role of semantic control mechanisms in creative thinking. The
aims of the first study were twofold: to explore the contributions of semantic knowledge and
control abilities to creative performance in divergent and convergent thinking tasks, beyond
domain-general cognitive mechanisms; and, to investigate age-related variations in semantic
and executive abilities, and their impact on creative thinking. The findings indicated that
divergent and convergent thinking rely on different aspects of semantic cognition. Specifically,
convergent thinking was found to be reliant on semantic knowledge, whereas divergent
thinking depended more on semantic control. This distinction was particularly evident in the
young group, suggesting that younger individuals may engage their semantic control
mechanisms more, while older people may rely on domain-general mechanisms of processing
speed and inhibition.
Building on these findings, the second study sought to further dissect the cognitive
mechanisms underlying creative thought. Specifically, the study investigated whether
performance on divergent and convergent thinking tasks could be explained by individual
differences in the ability to generate and evaluate ideas. A detailed analysis of the idea
generation process was conducted to understand how different retrieval strategies relate to
creative performance, with a focus on the role of clustering (retrieving related items) and
switching (transitioning between different clusters) in semantic search. The results indicated
that ability to generate multiple ideas enhances creative fluency, but strict evaluation may
hinder the originality of responses. Moreover, the findings suggested that multiple retrieval
strategies influence creative performance. Broad sampling of the semantic space is generally

4
beneficial for creative success, but the optimal strategy involves balancing exploitation of
specific areas of the semantic space and exploration of new areas to maximize creative output.
The third study explored the role of semantic control in a more cognitively complex
and ecologically valid realm: the production of extended naturalistic speech. Effective and
interesting discourse requires the generation of diverse ideas, but also crucially depends on the
ability to organise these ideas logically into a coherent narrative. Previous work has established
the role of various brain networks, including the Default Mode Network, Executive Control
Network and Semantic Control Network in supporting these processes, but the interplay
between them in spontaneous speech has been largely unexplored. To this end, the third study
investigated how creativity and coherence are related at the behavioural level, and the neural
mechanisms supporting these aspects of communication. A negative association between the
two constructs was found at the behavioural level, highlighting the cognitive balance required
to produce speech that is both interesting and comprehensible. Network activation analyses
indicated that more coherent speech was related to increased engagement of the Multiple
Demand Network, supporting the role of cognitive control processes in the maintenance of
coherent speech. Moreover, functional connectivity analyses indicated that divergence in
speech was related to increased functional connectivity between the default and executive
networks, aligning with previous work demonstrating default-executive coupling during
creativity tasks.
Collectively, the results of these studies highlight the multifaceted role of semantic
control in creativity, demonstrating that it is a critical cognitive mechanism that is differentially
engaged to support the generation, evaluation, and organization of original ideas.

5
 Lay Summary
Creative thinking is woven into the fabric of our lives: we often find ourselves in
situations where we must think of a quick fix for a problem, like repurposing a household item,
devising a new approach to tackle a task at work, or even putting together an elaborate
Halloween costume. But what does it mean to think creatively, and how do we do it? At the
heart of creative thinking lies our semantic memory, the extensive repository of knowledge
about the world that we accumulate over time. This knowledge serves as the building blocks
from which we create new ideas. But simply possessing the relevant knowledge is not enough,
we also need to be able to access and use the information effectively. This requires cognitive
processes known as semantic control mechanisms, which help us manage and apply what we
know in flexible and innovative ways.
While creativity is often seen as a mysterious, spontaneous process that occurs with
little conscious effort, recent work has highlighted the importance of exerting control over this
process. This thesis investigates how our ability to effectively use knowledge influences
different aspects of creative thinking.
The first study used behavioural measures to understand how the content of what we
know and our ability to use it flexibly contribute to creativity, and further, whether younger
and older adults rely on similar processes in pursuit of creative ideas. The findings showed that
younger and older adults were equally good at coming up with multiple creative ideas that were
both original and effective. However, while younger adults tend to rely more on their semantic
control abilities to come up with these ideas, this was less important for older adults. Instead,
older adults who were faster at solving problems, or better at avoiding stereotypical ideas, were
able to think in more creative ways. Interestingly, older adults were better at tasks that required
creative problem solving, likely because they could draw on their many years of knowledge
and experience. This suggests that while younger and older people might use different
strategies to achieve creative outcomes, both can be effective in their own ways.
Building on this, the second study examined how people come up with and judge ideas
when they are given problems with multiple clues. This study also looked at whether certain
strategies are more effective when searching through semantic memory, whether it is more
useful to explore a single area deeply, or to consider a broader range of possibilities. The results
suggested that multiple routes can lead to creative success, but they highlight the need to
balance focused exploration with broad thinking.

6
 Finally, the third study broke away from traditional ways of measuring divergent and
convergent thinking, instead examining how creativity emerges in everyday speech. Using
brain scans (functional Magnetic Resonance Imaging), the study found that people use different
brain networks when speaking creatively, compared to when their speech is more focused and
streamlined. Specifically, when the speaker is more creative, brain regions involved in
daydreaming and mental simulation, are more active, alongside areas related to the organisation
and management of ideas. On the other hand, when speech is closely tied to the topic, it
primarily relies on brain regions that organise and control thoughts.
The results from these studies establish that the ability to control and direct our
knowledge is important for creative thought. Semantic control mechanisms allow us to navigate
through the expanses of our semantic systems, avoiding stereotypical ideas and strategically
searching for less conventional alternatives, and selecting the most appropriate ideas. Overall,
this understanding can help to de-mystify creativity, showing that it involves both spontaneous
thinking and deliberate processes to help guide and refine our ideas.

7
 Acknowledgments
This PhD has been a long time coming – but here I am, writing the acknowledgements,
at the very end of this process. It is hard to put into words how grateful I am to the many people
that have supported me in this journey, and how lucky I feel to have had this incredibly
transformative experience.
We shall begin with my supervisors. To my primary supervisor, Dr. Paul Hoffman,
without whom I literally, figuratively, and metaphorically would not be standing at this finish
line today: Paul, apart from being simply brilliant, you have taught me so much about what it
means to be a good researcher. Thank you for your patience, for your invaluable feedback, and
for instilling in me the good practice of keeping a running document with comments. To my
second supervisor, Prof Sarah E. MacPherson: thank you Sarah, for all your wisdom and
guidance. Your insights, words of encouragement, and advice have been invaluable on this
journey. They say that good supervisors are hard to come by, but somehow, I managed to get
two of the best – and it has been my utmost pleasure to work with both of you.

Looking back, the PhD has been a wonderful (albeit, challenging) experience, one that
moulded me into not just a better researcher, but a better person. This is in no small part due to
the excellent people I have had the pleasure of meeting in Edinburgh. To all my delightfully
kooky yet brilliant friends at 7 George Square, and in Edinburgh: thank you for the laughter
and for the commiseration, I needed them both. Special mention to Esperanza (for holding me
through tears, salseo, and a covid booster), Greta and Fra (for always feeling like home),
Bérengère (for being my unwavering cinema/pastry buddy), Abby (for all the prosecco heists),
Yavor (for bringing the spice), and Sam (for all our time in the trenches). Thank you also to
Raf, Jolene (and Muggins!), JP, the wonderful Stats team of Jo, Emma & Umberto (and all the
tutors I’ve had a pleasure of working with over the years).

To my oldest friends back home, Tejal, Richard, Nada, Nain: thanks for being my life
support. To my family: suffice to say, this wouldn’t have been possible without you. Mama:
thank you for (ever so gently) nudging me in this direction, and sorry for abandoning you with
all the cats. Pops: thanks for your unwavering and unquestioning support through it all. To my
grandparents: I’m so happy that you get to see me complete this, and that I get to make you
proud. And finally, to my partner, Aju: thank you for everything, from all my 1000
personalities.

8
 Table of Contents
Signed Declaration ..................................................................................................................... 3

Abstract ...................................................................................................................................... 4

Lay Summary ............................................................................................................................. 6

Acknowledgments...................................................................................................................... 8

Table of Contents ....................................................................................................................... 9

Table of Figures ....................................................................................................................... 13

Table of Tables ........................................................................................................................ 14

List of Abbreviations ............................................................................................................... 16

Chapter 1 .................................................................................................................................. 17

Semantic Cognition .................................................................................................................. 17

1.1 Introduction ........................................................................................................... 17

1.2 Controlled Semantic Cognition ................................................................................ 18
1.3. Hub and Spoke Model............................................................................................. 18
1.3.1. Evidence from Neuropsychology: Semantic Dementia.......................................................... 19
1.3.2 Evidence from Neuroimaging Studies .................................................................................... 21
1.3.3. To the ATL and beyond .......................................................................................................... 22
1.3.4. Interim Summary .................................................................................................................. 24

1.4. Semantic Control ................................................................................................... 24

1.4.1. Evidence from Neuropsychology: Semantic Aphasia ............................................................ 25
1.4.2. Evidence from Neuroimaging Studies ................................................................................... 27
1.4.3. Controlled Retrieval and Semantic Selection ........................................................................ 27
1.4.4. More than meets the I(FG) .................................................................................................... 28
1.4.5. The interaction between representation and control ............................................................. 29

1.5. Chapter Summary .................................................................................................. 31

Chapter 2 .................................................................................................................................. 32

Creativity.................................................................................................................................. 32

2.1. Definitions............................................................................................................. 32
2.1.1. Product vs Process ............................................................................................................... 32
2.1.2. Divergent and Convergent Thinking ....................................................................................... 33

9
 2.2. Situating Semantic Memory within Creativity Theories ............................................... 34
2.3. Associative Theories ............................................................................................... 35
2.3.1. Evidence from Computational Network Studies.................................................................... 36
2.3.2. Individual Differences Approach ........................................................................................... 36

2.4. Executive Accounts of Creative Cognition ................................................................. 38

2.4.1. Executive Functions.............................................................................................................. 38
2.4.2. Fluid Intelligence .................................................................................................................. 39
2.4.3. A note on the benefits of reduced control ............................................................................. 40
2.4.4. Free Association vs Goal Directed Association ..................................................................... 41
2.4.5. Semantic Control ................................................................................................................. 43

2.5. Dual Process Models .............................................................................................. 44

2.6. Neural basis of creative thinking .............................................................................. 45
2.6.1. Associative processes .......................................................................................................... 45
2.6.2. Executive Processes ............................................................................................................. 46
2.6.3. Interplay between the default and executive networks .......................................................... 47
2.6.4. Semantic Control Network ................................................................................................... 48

2.7. Thesis Roadmap .................................................................................................... 50

Chapter 3 .................................................................................................................................. 52

Investigating age related differences in semantic control mechanisms involved in creative

cognition .................................................................................................................................. 52

3.1. Introduction .......................................................................................................... 54

3.2. Methods................................................................................................................ 59
3.2.1 Transparency and Openness.................................................................................................. 59
3.2.2. Participants .......................................................................................................................... 59
3.2.3. Procedure ............................................................................................................................. 60
3.2.4. Measures of Intelligence ....................................................................................................... 60
3.2.5. Measures of Executive Functioning ....................................................................................... 62
3.2.6. Measures of Semantic Ability ................................................................................................ 64
3.2.7. Measures of Creativity: Self Report ....................................................................................... 65
3.2.8. Measures of Creativity: Convergent and Divergent Thinking Tasks ......................................... 67
3.2.9. Scoring the Alternate Uses Task ............................................................................................ 68
3.2.10. Statistical Analyses ............................................................................................................ 70

3.3. Results.................................................................................................................. 71
3.3.1. Age differences on measures of divergent thinking, semantic knowledge, and control
mechanisms .................................................................................................................................. 71
3.3.2. Relationships between creativity, semantic and executive abilities ....................................... 74
3.3.3. Predictors of performance on RAT ......................................................................................... 76
3.3.4. Predictors of performance on AUT ........................................................................................ 77

3.4. Discussion ............................................................................................................ 82

10
Chapter 4 .................................................................................................................................. 88

Dissecting the RAT: Exploring the Dynamics of Semantic Search in Creative Thinking ...... 88

4.1. Introduction .......................................................................................................... 89

4.2. Methods................................................................................................................ 93
4.2.1. Participants .......................................................................................................................... 93
4.2.2. Procedure ............................................................................................................................. 94
4.2.3. Tasks .................................................................................................................................... 94
4.2.4. Measures of Clustering/Switching Behaviour ........................................................................ 97
4.2.5. Analysis Strategy .................................................................................................................. 99

4.3. Results................................................................................................................ 100

4.3.1. Relationships between Cognitive Measures and Creative Performance .............................. 100
4.3.2. Generation and Evaluation as Predictors of Creative Thinking ............................................. 103
4.3.3. Principal Components Analysis .......................................................................................... 103
4.3.4. Semantic search components as predictors of Standard RAT performance ........................ 104
4.3.5. Semantic search components as predictors of AUT performance ....................................... 107

4.4. Discussion .......................................................................................................... 109

4.4.1. How do generation and evaluation predict creative performance? ...................................... 110
4.4.2 How do retrieval strategies influence creative performance? ............................................... 112

Chapter 5 ................................................................................................................................ 116

Balancing Act: A Trade-off Between Coherence and Creativity in Spontaneous Speech .... 116

5.1. Introduction ........................................................................................................ 118

5.2. Study 1: Methods ................................................................................................. 125
5.2.1. Participants ........................................................................................................................ 125
5.2.2. Design and Procedure ........................................................................................................ 126
5.2.3. Measures ............................................................................................................................ 127
5.2.4. Analysis Strategy ................................................................................................................ 129

5.3. Study 1: Results ................................................................................................... 130

5.3.1. Descriptive and correlational analysis of speech metrics ................................................... 130
5.3.2. Global Coherence as a Predictor of Semantic Divergence in Speech .................................. 132
5.3.3. Exploratory Analyses of the impact of Task Instructions on Speech Metrics ........................ 133

5.4. Study 1: Discussion .............................................................................................. 135

5.5. Study 2: Methods ................................................................................................. 135
5.5.1. Participants ........................................................................................................................ 136
5.5.2. Design and Procedure ........................................................................................................ 136
5.5.3. Materials and Measures ...................................................................................................... 136
5.5.4. Processing of speech samples............................................................................................ 137
5.5.5. Image Acquisition and Processing ...................................................................................... 138
5.5.6. Analyses ............................................................................................................................. 139

5.6. Study 2: Results ................................................................................................... 142

5.6.1. Whole Brain Analyses ......................................................................................................... 142
5.6.2. Network Activation Analyses .............................................................................................. 145

11
 5.6.3. Functional Connectivity Analyses ....................................................................................... 147

5.7. General Discussion .............................................................................................. 150

Chapter 6 ................................................................................................................................ 158

General Discussion ................................................................................................................ 158

6.1. Chapter Summaries ............................................................................................. 158

6.2. Broader Theoretical Implications ........................................................................... 161
6.2.1. Controlled Semantic Cognition........................................................................................... 161
6.2.2. Dual Process Theories ........................................................................................................ 162
6.2.3. Strategic Search ................................................................................................................. 163
6.2.4. Extending the Dual Process: MemiC Framework ................................................................. 164
6.2.5. Key Distinction between New and Old Ideas: The Episodic vs Semantic Question .............. 165
6.2.6. Segmenting the Gyrus: The Distinct Roles of BA 45 and BA 47 in Semantic Control ............. 167
6.2.7. Bringing Creativity into the Real World ................................................................................ 168

6.3. Conclusion .......................................................................................................... 168

References .............................................................................................................................. 170

Appendices ............................................................................................................................. 232

Appendix A: Supplemental Materials (Chapter 3) ................................................................ 232

Appendix B: Supplemental Materials (Chapter 4)................................................................. 236

Appendix C: Supplemental Materials (Chapter 5)................................................................. 238

12
 Table of Figures

Chapter 1 .................................................................................................................................. 17

Figure 1: Masks of Networks of Interest. ......................................................................... 30

Chapter 3 .................................................................................................................................. 52

Figure 1: Pearson correlations between creativity, semantic and executive abilities in the
combined dataset. ....................................................................................................... 74
Figure 2: Two-way interaction plot of the predicted effects of semantic selection on AUT
Originality scores. ........................................................................................................ 78
Figure 3: Effects of cognitive abilities on AUT Ratings. ...................................................... 80

Chapter 4 .................................................................................................................................. 88

Figure 1: Schematic Illustration of Tasks. ........................................................................ 97

Figure 2: Pearson correlations between measured abilities. ........................................... 102
Figure 3: Predictors of Standard RAT Performance. ........................................................ 106
Figure 4: Predictors of AUT Performance....................................................................... 109

Chapter 5 ................................................................................................................................ 116

Figure 1: Boxplots of DSI and Coherence Across Datasets and Ages. .............................. 130
Figure 2: DSI and Coherence Across Instruction Conditions. .......................................... 133
Figure 3: Workflow from Data Collection to Neural Analysis. .......................................... 137
Figure 4: Masks of the networks of interest. .................................................................. 140
Figure 5: Whole brain activation for semantic compared to baseline speech. ................... 143
Figure 6: Unthresholded maps (beta values) of the effects of the regressors on neural
activation. ................................................................................................................. 143
Figure 7: Effects of regressors in each network. ............................................................. 147
Figure 8: Effects of regressors on network connectivity. ................................................. 150

13
 Table of Tables
Chapter 3 .................................................................................................................................. 52

Table 1: Descriptive Statistics by Age Group.................................................................... 73

Table 2: Principal Components Analysis of creativity measures. ....................................... 75
Table 3: Effects of age, semantic and executive functions on Remote Associates Task (RAT)
performance. ............................................................................................................... 77
Table 4: Analyses of effects of age, semantic and executive functions on the Alternate Uses
Task (AUT) Originality. ................................................................................................... 78
Table 5: Analyses of effects of age, semantic and executive functions on Alternate Uses Task
(AUT) Ratings. .............................................................................................................. 79
Table 6: Analyses of effects of age, semantic and executive functions on Alternate Uses Task
(AUT) Uniqueness. ........................................................................................................ 81
Table 7: Analyses of effects of age, semantic and executive functions on Alternate Uses Task
(AUT) Fluency. .............................................................................................................. 81

Chapter 4 .................................................................................................................................. 88

Table 1: Descriptive Statistics for Cognitive Measures. ................................................... 101

Table 2: Effects of generation and evaluation abilities on creative performance. ............... 103
Table 3: Exploratory Principal Components Analysis of clustering/switching metrics. ....... 104
Table 4: Effects of the PCA components on Standard RAT performance. .......................... 105
Table 5: Effects of mean clustering and switching on Standard RAT performance. ............ 107
Table 6: Effects of PCA-derived components on AUT Metrics. ......................................... 108

Chapter 5 ................................................................................................................................ 116

Table 1: Demographic summary of datasets. ................................................................ 126

Table 2: Descriptive Statistics for all measures.............................................................. 131
Table 3: Correlation between measures in each individual sample and in the combined
dataset. ..................................................................................................................... 132
Table 4: Principal Components Analysis of Metrics. ....................................................... 132
Table 5: Effects of Coherence, Word Count, and Age on DSI. .......................................... 133
Table 6: Effects of Coherence, Number of Words and Condition on DSI. .......................... 134
Table 7: One-sample t-tests for effects across each ICA-derived network. ....................... 145

14
Table 8: Paired t-tests for Network Differences Across Semantic and Coherence Measures
................................................................................................................................. 146
Table 9: One sample t-tests for effects in each network pair. .......................................... 148
Table 10: Paired t-tests for Network Differences in Semantic, Coherence, DSI, and Word
Count Measures. ........................................................................................................ 149

15
 List of Abbreviations
AG = Angular Gyrus
ALE = Activation Likelihood Estimation
ATL = Anterior Temporal Lobe
AUT = Alternate Uses Task
CBI = Creative Behaviour Inventory
CSC = Controlled Semantic Cognition
E-RAT = Evaluation RAT
DLPFC = Dorsolateral Prefrontal Cortex
DMN = Default Mode Network
DSI = Divergent Semantic Integration
fMRI = Functional Magnetic Resonance Imaging
GC = Global Coherence
G-RAT = Generation RAT
ICAA = Inventory of Creative Activities and Achievements
IFG = Inferior Frontal Gyrus
IPL = Inferior Parietal Lobe
LIFG = Left Inferior Frontal Gyrus
LSA = Latent Semantic Analysis
MEG = Magnetoencephalography
MDN = Multiple Demand Network
mPFC = Medial Prefrontal Cortex
PET = Positron Emission Tomography
PCC = Posterior Cingulate Cortex
PFC = Pre-frontal Cortex
pMTG = Posterior Middle Temporal Gyrus
RAT = Remote Associates Task
SA = Semantic Aphasia
SCN = Semantic Control Network
SD = Semantic Dementia
SMG = Superior Middle Gyrus
TMS = Transcranial Magnetic Stimulation
VLPFC = Ventrolateral Pre-frontal Cortex

16
 Chapter 1

Semantic Cognition
1.1 Introduction

As we interact with the world around us, we accumulate an extensive amount of

semantic knowledge, information about facts and meanings of concepts (Lambon Ralph et al.,
2017; Patterson et al., 2007; Tulving, 2002). Typically, we use this knowledge effortlessly to
guide our behaviour, navigate everyday environments, and communicate with others. But, there
are times where we must draw on this knowledge in more deliberate, effortful ways. For
example, consider a situation in which you suddenly hear someone yell the word ‘duck’. You
may quickly gather that the speaker is not overly excited about a seeing a duck (in the context
of the bird that quacks) but is warning you to take cover (perhaps from a stray tomato flying in
your general direction). Semantic control mechanisms (i.e., processes that regulate the efficient
retrieval of knowledge) may be particularly useful here. These mechanisms allow us to access
and flexibly manipulate information to meet specific task demands, filtering out irrelevant but
strongly ingrained information while guiding our search toward ideas that are less immediate,
but more suitable (Badre et al., 2005; Lambon Ralph et al., 2017). While the importance of
these specific control mechanisms to semantic cognition and everyday language use has been
established, their influence in more complex cognitive domains remains less explored.
One such area is creative thinking, which involves generating ideas that are both
original and appropriate. To do this, one must avoid common ideas, and instead, resume a
search for more unconventional links between concepts (Beaty & Kenett, 2023; Benedek et al.,
2023; Green et al., 2023). Consequently, semantic control mechanisms might be crucial in
guiding the retrieval and selection of ideas that meet the specific criteria of creative tasks.
The aim of this thesis is to investigate how the different facets of semantic control
contribute to creativity. Specifically, this work will examine how the ability to retrieve,
manipulate and select relevant semantic knowledge influences our capacity to produce original
ideas. To provide a theoretical foundation for this work, I will review current neurocognitive
models of semantic cognition in Chapter 1, examining how semantic knowledge is stored and
accessed across different contexts. Following this, I will turn to the creativity literature in
Chapter 2, providing an overview of current theories regarding the roles of semantic memory
and executive control processes in creative thinking.

17
1.2 Controlled Semantic Cognition
Semantic cognition refers to the set of neurocognitive processes that are involved in
storing, organising, and using knowledge about the world (Lambon Ralph et al., 2017;
Patterson et al., 2007; Tulving, 2002). This knowledge, encompassing both verbal and non-
verbal information, allows us to interpret meaning, recognise and use objects, and engage in
discourse. The Controlled Semantic Cognition (CSC; Lambon Ralph et al., 2017) framework
proposes that this ability to comprehend meaning is underpinned by two separate yet interacting
systems. First, the semantic representation system that stores our conceptual knowledge,
learned through multimodal experiences with world around us. Second, the semantic control
system that enables the flexible use of this knowledge according to task requirements. In the
following sections, I will outline how these two systems operate at the cognitive and neural
levels. First, I will discuss the semantic representation system, focusing on the hub-and-spoke
model (Patterson et al., 2007), and present supporting evidence. Following this, I will examine
the role of the semantic control system and how it interacts with the representational system to
guide behaviour.

1.3. Hub and Spoke Model

Throughout our existence, we interact with the world around us, extracting knowledge
about a vast range of subjects - from objects, emotions, and abstract concepts to the culinary
versatility of apples in both sweet and savoury dishes. But what does it mean to know about
apples, and how do we come to have this knowledge? Embodied views suggest that we learn
about objects through multiple interactions with them, extracting a vast amount of information
about their flavour, feel and form (Meteyard et al., 2012). For example, biting into a crisp apple
provides immediate sensory information: the redness, the crunch, the juicy sweetness. Through
multiple exposures to apples across different contexts, we might additionally learn that they
come in different flavours and colours (like the green Granny Smith apples), the vitamins they
contain, and that a jealous stepmother once tried to get rid of Snow White with a poisoned
apple. We store all this information related to apples, activating a task-relevant subset of this
knowledge at any given time, the scope of which can vary from specific context-relevant
features to an overarching subset of knowledge.
The question of where and how this knowledge is represented in the brain has been of
paramount interest in the literature. The classical Wernicke-Meynert model proposed a purely
distributed approach, where representations are encoded across interconnected modality-
specific association cortices, that reflect the underlying sensory-motor experiences (Eggert &

18
Wernicke, 1977). This idea of a distributed, experientially driven semantic system is echoed in
strongly embodied accounts of semantic cognition (Martin, 2007; Pulvermüller, 2005;
Thompson-Schill, 2003). While these modality-specific areas are crucial to the semantic
system, they are not sufficient: there is a need for a component of the system that can identify
and enable generalization across similar concepts. To meet this need, the distributed view was
extended by Damasio (1989) who proposed the existence of various convergence zones, where
different types of information can be integrated into abstract, multimodal representations.
The hub-and-spoke theory (Patterson et al., 2007; Patterson & Lambon Ralph, 2016)
developed this idea further, suggesting the existence of a single convergence zone that supports
higher order abstract generalisation of knowledge. This theory proposes that knowledge is
stored in a distributed fashion across the brain, with visual and perceptual attributes encoded
and stored in modality specific brain areas or ‘spokes’. For example, information about the
redness of apples in visual regions, the crunchiness in somatosensory regions, and the sweet
taste in the gustatory areas. Crucially, however, these spoke regions are interconnected by a
central transmodal ‘hub’ region, which integrates diverse sensory information into a cohesive
concept of ‘apple’. The model further suggests that this integration takes place in the anterior
temporal lobes (ATL), positioning this region as a central hub for semantic processing,
consistent over time and modalities, integrating information from the various sensory-motor
spokes (Lambon Ralph et al., 2010). The location of the ATL makes it well suited for this
position, as it is connected to primary sensory regions, as well as medial and frontal areas. In
the following sections, I will review converging evidence from neuropsychological and
neuroscientific work that supports this view.

1.3.1. Evidence from Neuropsychology: Semantic Dementia

Compelling evidence for this ATL-hub view comes from neuropsychological work on
semantic dementia, a variant of frontotemporal dementia characterized by the progressive
neurodegeneration of the anterior temporal lobes (Hodges et al., 1992; Hodges & Patterson,
2007; Patterson et al., 2007). While initially localized to the ATLs, the atrophy in this condition
spreads to other areas of the temporal and inferior frontal cortices as the disease progresses.
Due to this ATL-focused atrophy, patients with semantic dementia exhibit impairments specific
to semantic knowledge, while maintaining relatively preserved abilities in other domains, e.g.,
visuospatial processing, episodic and autobiographical memory, and executive control (Hodges
& Patterson, 2007; Snowden et al., 2001).

19
 Multiple clinical features of semantic dementia support the idea of the ATL as a central
representational hub (Lambon Ralph et al., 2010a; Patterson et al., 2007). First, patients
experience a progressive erosion of knowledge, demonstrated across a range of tasks,
modalities, and word types. Patients exhibit semantic impairments in expressive (e.g., picture
naming, generating verbal descriptions), receptive (e.g., word comprehension), and non-verbal
tasks (e.g., using objects based on properties, picture matching) as well as across various
sensory modalities (Bozeat et al., 2000, 2002; Coccia et al., 2004; Hoffman & Lambon Ralph,
2011; Ikeda et al., 2006; Jefferies et al., 2009a). Second, the amount of atrophy and
hypometabolism in the anterior temporal lobes is related to the degree of degradation of
semantic memory, quantified by performance on a range of semantic tasks (Mion et al., 2010).
Moreover, focal bilateral atrophy is accompanied by reduced connectivity between the anterior
temporal lobes and modality specific regions, correlating with the degree of deficits on
neuropsychological tasks (Guo et al., 2013). Collectively, these findings support the necessity
of both the central hub and the spoke regions to the architecture of the semantic system. Chen
et al. (2019) further reinforced the hub-and-spoke model in a sample of semantic dementia
patients, identifying the left anterior fusiform gyrus as an amodal semantic hub region,
alongside modality-specific regions in the left and right temporal areas that support verbal and
nonverbal semantic processing, respectively.
Third, there is a systematic and progressive degradation of knowledge from specific to
general information, paralleling the spread of atrophy across the anterior temporal lobes
(Hodges et al., 1995; Patterson et al., 2006, 2007; Rogers et al., 2004, 2015; Warrington, 1975;
Woollams et al., 2008). In the initial stages, patients show preserved performance for familiar,
typical, and general concepts over unfamiliar, atypical, and specific information. For example,
patients will retain basic category exemplars (‘pie’, a common dessert item) while losing
information about specific or atypical examples (‘cronut’, an unusual - and delicious - dessert
item). This pattern indicates that concepts that are encountered early and more frequently, or
are representative category exemplars, are more strongly encoded and resistant to damage. As
the hub deteriorates, fine-grained details are lost first, followed by the degradation of more
general representations, leading to broader semantic deficits (Hoffman et al., 2012). Finally,
semantic dementia patients make consistent errors with concepts across different modalities
and tasks, indicating a general breakdown in their ability to integrate semantic information
(Jefferies & Lambon Ralph, 2006). Taken together, the evidence presented in this section
suggests that semantic dementia results in the degradation of a central hub area, located in the
anterior temporal lobes.

20
 The critical role of the anterior temporal lobes to semantic representation is also evident
when considering the pattern of deficits exhibited in other neurological conditions associated
with semantic impairments. For instance, while there is some semantic impairment in
Alzheimer’s Disease (AD), this is less severe and occurs later, as compared to the primary
episodic memory deficits (Giffard, 2001; Grossman, 2003). Moreover, the hypometabolism in
AD is widespread but the anterior temporal lobes remain relatively spared; compared to the
focal temporal lobe atrophy seen in semantic dementia (Galton, 2000). Further, studies of
acquired damage to the anterior temporal lobes, such as from temporal lobe epilepsy, indicate
that unilateral damage results in milder semantic impairments (Lambon Ralph et al., 2012).
Finally, there are category-specific variations where different types of knowledge are
selectively impaired by neurological disorders. For example, patients with lesions to spoke
regions have modality-/domain-specific impairments, e.g., auditory-verbal deficits seen in
Wernicke’s Aphasia (Thompson et al., 2015); visual object naming deficits seen in optic
aphasia (Plaut, 2002); category specific deficits in knowledge about natural vs man-made items
seen in patients with herpes simplex virus encephalitis (Lambon Ralph et al., 2006). These
findings support the connectivity constrained hub and spoke model, which can explain both
general and category specific deficits, highlighting how damage to either the hub or spoke
regions can lead to distinct semantic deficits.

1.3.2 Evidence from Neuroimaging Studies

The hub-and-spoke model has been validated and extended via convergent evidence
from a range of empirical studies with neurologically healthy patients, using positron emission
tomography (PET; e.g., Devlin et al., 2000; Rogers et al., 2006), functional Magnetic
Resonance Imaging (fMRI; e.g., Binder et al., 2009; Visser et al., 2010) and
magnetoencephalography (MEG; e.g., Marinkovic et al., 2003). Further, causal evidence for
the ATL as a representational hub region comes from studies applying transcranial magnetic
stimulation (TMS) over the ATL to simulate impairments in semantic dementia (Binney et al.,
2010; Lambon Ralph & Pobric, 2009; Pobric et al., 2007, 2009, 2010a, 2010b). As the location
of the ATL makes this region prone to susceptibility effects and signal drop-out, early fMRI
studies either reported inconsistent effects or failed to find activation in the ATL (Binder et al.,
2009; Devlin et al., 2000; Visser et al., 2010). Using distortion-corrected fMRI to ameliorate
the signal dropout, more recent neuroimaging studies have demonstrated ATL activation in
response to a wide range of multimodal semantic tasks, bolstering the view of the ATL as a
pan-modal hub (Visser et al., 2012; Visser & Lambon Ralph, 2011). Moreover, both similarity

21
and associative semantic judgments engage the ATL, indicating that this region is involved in
processing different types of semantic relationships (Jackson et al., 2015). Finally, imaging
studies have been able to locate the cross-modal hub region more precisely within the
ventrolateral ATL, as this region is consistently engaged in semantic processing, irrespective
of modality or semantic categories (Binney et al., 2010; Visser et al., 2012; Visser & Lambon
Ralph, 2011).
A key characteristic of a coherent representational system is the ability to bring together
many different sources of information, and to be able to generalise based on semantic
similarities (Lambon Ralph et al., 2010b). In line with this, whole brain multivoxel searchlight
analyses indicated that the ATL encodes conceptual properties of objects, such as location and
action, independent of their perceptual properties (Peelen & Caramazza, 2012). This suggests
that the ATL abstracts information away from the perceptual level to form abstract higher order
semantic representations, making it a good candidate for a representational hub. The location
and connectivity of the ATL support this role, with structural connectivity studies indicating
that it demonstrates intrinsic resting-state connectivity with modality-specific areas (Guo et al.,
2013). Further, an Activation Likelihood Estimation (ALE) metanalysis of 97 neuroimaging
studies found functionally graded specialization within the ATL, e.g., verbal stimuli tend to
activate more dorsal ATL regions, which are connected to auditory and speech areas (Rice et
al., 2015). This suggests that different subregions of the ATL are preferentially activated
depending on the type of information being processed, arising from the differential connectivity
of these subregions with various sensory-motor and limbic regions. This pattern of connectivity
allows the ATL to integrate diverse types of sensory information into unified semantic
representations that are coherent and contextually relevant (Coutanche & Thompson-Schill,
2015). Moreover, the ATL dynamically alters its functional connectivity with spoke regions,
depending on the modality of the information being processed. For instance, Chiou and
Lambon Ralph (2019) found that while the ATL itself was consistently engaged in processing
both action and place knowledge, it increased its functional connectivity with frontoparietal
regions for action information, and with medial regions for place information.

1.3.3. To the ATL and beyond

More broadly, semantic processing engages a distributed set of multimodal regions: the
inferior parietal lobe (including the angular gyrus and supramarginal gyrus), middle temporal
gyrus, ventral temporal cortex, ventromedial prefrontal cortex, and posterior cingulate gyrus
(Binder et al., 2009). Within this network, ‘spoke’ regions play a crucial role in semantic

22
cognition by processing specific types of sensory-motor information. For example, while
stimulation of the ATL results in generalised deficits, i.e., slower object naming for both living
and non-living things; stimulation of the inferior parietal areas, that are typically involved with
action perception, results in impaired naming specifically for non-living, manipulable objects
(Pobric et al., 2010b). This category-specific effect highlights the importance of both the hub
and spoke regions in processing modality specific semantic information. Moreover, Mollo et
al. (2017) demonstrated a dissociation between visual and motor regions, with visual regions
responding more to animals and motor regions to manmade objects. These results further
highlight the differential engagement of sensory-motor features by spoke regions and their
essential contribution to semantic cognition.
The broader semantic network also overlaps with the Default Mode Network (DMN),
a set of regions associated with spontaneous internal thought processes, such as mind-
wandering, episodic retrieval and future thinking (Andrews-Hanna, 2012; Fox et al., 2015;
Smallwood et al., 2012). The DMN, which includes regions such as the angular gyrus (AG),
posterior cingulate cortex (PCC), medial prefrontal cortex (mPFC), and medial temporal gyrus
(MTG), typically deactivates during externally focused or cognitively demanding tasks,
reflecting its primary role in internally directed cognition.
Within this network, the AG plays a particularly significant role in semantic cognition.
The AG deactivates in response to both semantic and non-semantic tasks, but the degree of
deactivation increases with task difficulty (Humphreys et al., 2015, 2021; Humphreys &
Lambon Ralph, 2017). This lack of specificity for semantic information suggests that the AG
may have a more generalized role in cognitive processing. Further, this pattern indicates that
the role of the AG is distinct from that of the ATL: the AG is not involved in storage of semantic
information, but instead serves a more dynamic function within higher-order cognitive systems,
such as those involving executive functions (Farahibozorg et al., 2022). Specifically, the AG
has been thought to act as an online multimodal buffer for incoming information, integrating
sensory-motor information and facilitating efficient retrieval (Kuhnke et al., 2023). This
buffering capacity allows the AG to support the combination of different types of information
(e.g., from different sensory modalities and spatiotemporal contexts) making it essential for
tasks that require the retrieval and manipulation of semantic knowledge (Humphreys et al.,
2021). Moreover, research suggests a functionally distinct role for the angular gyrus in
facilitating the retrieval of strong associations (Davey et al., 2015, 2016), making it particularly
important for retrieving well-established and contextually-relevant information. It is important
to note that the role of the AG extends beyond semantics to include episodic memory (i.e.,

23
memories about specific events we have experienced), where it similarly functions as an
dynamic online buffer (Humphreys et al., 2021). Together, this evidence highlights the broader
role of the AG in supporting complex cognitive operations as well as semantics.

1.3.4. Interim Summary

Taken together, multiple strands of evidence reliably demonstrate the existence of a hub-
and-spoke system in semantic cognition; indicating that semantic representations are
distributed across the brain, with modality specific neural information activated during
semantic processing. However, only a subset of this information is relevant for a given context
at a time. Concepts are not static and invariant, but change according to the context, including
current task goals, individual differences, and recent or long-term experiences (Yee &
Thompson-Schill, 2016). Different aspects of the semantic representation may be more or less
salient in different contexts (Hoenig et al., 2008; Kiefer & Pulvermüller, 2012), requiring
varying degrees of control. Returning to our apple example, using knowledge in a routine or
familiar context does not require control (apple-computer); but, if context accentuates atypical
usage of semantic information (apple-evil stepmother), more control is needed. Thus, semantic
representations are accessed in a dynamic way, with various perceptual features being accessed
at different timepoints, reflecting a flexible and fluid conceptualisation of semantic memory
that can change as a function of task. This dynamic and context-sensitive nature of semantic
representations necessitates robust mechanisms of semantic control, which will be discussed
in the next section.

1.4. Semantic Control

In the above sections, I introduced the notion that semantic representations have many
features and associations that are not always relevant to the current context and may even be
distracting. Semantic control refers to the ability to flexibly access, manipulate and select those
aspects of a concept that are relevant to the task context (Badre et al., 2005; Badre & Wagner,
2007; Lambon Ralph et al., 2017b; Reilly et al., 2023). This includes suppressing dominant
aspects of meaning in favour of less frequent word meanings, flexibly shifting between tasks,
and resolving competition when multiple representations are activated. The investigation into
semantic control has its roots in neuropsychology, stemming from a dissociation between
semantic dementia patients who struggled with loss of knowledge (Warrington, 1975), and
patients with ‘semantic access’ problems, who possessed the relevant semantic knowledge but
their access to this knowledge was inconsistent or inefficient (Campanella et al., 2009;

24
Warrington & Shallice, 1979). Contemporary work with semantic aphasia patients who have
left hemispheric damage to frontoparietal or posterior temporal areas (Thompson et al., 2022),
or using experimental tasks to manipulate control demands, have also established the
importance of semantic control mechanisms to semantic cognition. In the following sections, I
will review the neuropsychological and neuroimaging evidence for the role of the semantic
control system.

1.4.1. Evidence from Neuropsychology: Semantic Aphasia

The term ‘semantic aphasia’ (SA) was first used by Head (1926) and Luria (1973) to
describe patients with multimodal semantic deficits following a left hemisphere stroke.
Semantic aphasia patients typically have damage to regions in the inferior frontal gyrus, but
recent neuropsychological work indicates that damage to either the prefrontal or
temporoparietal cortex can result in equivalent control deficits, highlighting the distributed
nature of the semantic control network (Thompson et al., 2022). This syndrome presents a
unique pattern of impairments compared to semantic dementia – rather than showing
degradation of conceptual knowledge, semantic aphasia is characterised by deficits in the
flexible retrieval and control of knowledge.
Several clinical features of semantic aphasia are relevant here. First, a key characteristic
is the inconsistency in response patterns: semantic aphasia patients generate highly variable
responses to the same concept within and across tasks, in contrast to semantic dementia
patients, who display strong within-item and across-task consistency (Gardner et al., 2012;
Jefferies et al., 2007, 2008; Jefferies & Lambon Ralph, 2006; Thompson et al., 2015). This
suggests that semantic aphasia patients can demonstrate knowledge in one context but not
another, highlighting their deficits in accessing knowledge. Second, while semantic dementia
patients show similar performance across semantic tasks regardless of the modality, the
performance of semantic aphasia patients varies as a function of the control requirements on
the task, with worse performance on tasks requiring more control (Corbett et al., 2009; Noonan
et al., 2010). For example, they struggle with retrieving the less common meanings of
ambiguous words (bank-money vs. bank-river), have issues with connecting concepts that are
further apart in the semantic space, and have trouble with inhibiting strong distractors.
Moreover, semantic aphasia patients are more impaired on thematic, rather than object,
knowledge; with worse performance on tasks tapping into retrieval of weak associations that
require higher levels of control (Thompson et al., 2017). For example, when asked to identify
the relationship between a “camel” and a “cactus” (a weak thematic association), these patients

25
struggle more compared to tasks requiring identification of a straightforward object-property
relationship, such as matching camel to the category of animals. Additionally, Montefinese et
al. (2021) demonstrated that semantic aphasia patients show a reduction in the ability to select
features of low semantic relevance, such as identifying non-dominant aspects of a concept (e.g.,
zebra-stripes rather than zebra-legs), highlighting the importance of control processes for
selection of atypical semantic associations.
Third, while semantic aphasia patients have trouble retrieving more weakly encoded
meanings (Noonan et al., 2010), they can respond accurately if ambiguous targets are cued
(Lanzoni et al., 2019). For instance, if the word "bank" is presented, semantic aphasia patients
may struggle to retrieve the less dominant meaning; but, if the target is cued with a picture of
a river, they are more likely to correctly identify the riverbank meaning, showing that
appropriate cues can guide them toward the correct interpretation even when it is not the most
immediately accessible. In fact, patients with semantic aphasia benefit strongly from external
cues, irrespective of the cue type, and are thrown off by miscues designed to increase activation
of strong competitors. Appropriate cues are most beneficial, suggesting that their semantic
knowledge is intact but cannot be retrieved without suitable external constraint. Fourth, unlike
semantic dementia, semantic aphasia patients demonstrate smaller effects of frequency and
familiarity, indicating that their performance deficits do not result from a specific-to-general
pattern of knowledge degradation, but an inability to flexibly retrieve and use semantic
information (Hoffman, Jefferies, et al., 2011; Hoffman, Rogers, et al., 2011; Jefferies et al.,
2009b; Jefferies & Lambon Ralph, 2006; Thompson et al., 2018). Again, these effects appear
inconsistently across tasks and conditions, further highlighting the significance of the control
deficits (Rogers et al., 2015b). When they do demonstrate frequency effects, they tend to be
more impaired with high-frequency words (as opposed to the low-frequency effect seen in
semantic dementia), as these words appears in more diverse contexts and impose greater
demands on impaired control processes (Hoffman, Rogers, et al., 2011). Additionally, the types
of errors made are also relevant: semantic aphasia patients make more associative errors (e.g.,
saying 'desert' instead of the correct specific label for an item found in a desert), implying a
failure of controlled retrieval wherein irrelevant associations may cause interference (Jefferies
& Lambon Ralph, 2006). These errors reflect difficulties in the suppression of strong but
irrelevant associations, leading to erroneous intrusions that suggest a struggle to overcome
competition from related concepts. This pattern highlights the core deficit in semantic aphasia:
a breakdown in the ability to flexibly and selectively retrieve the most relevant aspects of
semantic knowledge, leading to errors driven by automatic yet inappropriate associations.

26
 Overall, the clinical features of semantic aphasia underscore the critical role of semantic
control in maintaining flexible and contextually appropriate access to semantic knowledge. The
pattern of impairments highlights the core deficit in semantic aphasia: rather than a loss of
knowledge, there is a disruption to the ability to dynamically retrieve and apply this knowledge
across different contexts.

1.4.2. Evidence from Neuroimaging Studies

Semantic control engages a distributed network, including the inferior frontal gyrus
(IFG), dorsomedial prefrontal cortex (dmPFC), posterior middle temporal gyrus (pMTG), and
anterior cingulate cortex (ACC; Jackson, 2021; Noonan et al., 2013). It is predominantly a left-
lateralised network, with significant involvement of left inferior frontal gyrus (LIFG) and
posterior temporal regions, as well as the bilateral dmPFC and right IFG to a lesser extent. This
network shows reliable activations during semantic control tasks regardless of stimulus
modality, confirming the multimodal nature of semantic control processes.

1.4.3. Controlled Retrieval and Semantic Selection

A notable contribution of the neuroimaging literature has been the teasing apart of different
semantic control mechanisms associated with anatomically separable subregions of the left
inferior frontal gyrus (Badre et al., 2005; Badre & Wagner, 2007; Thompson-Schill et al.,
1997). Specifically, the literature has identified two specific mechanisms: controlled retrieval
and semantic selection. Controlled retrieval refers to the top-down, goal-directed retrieval of
semantic information that occurs when bottom-up stimulus driven activation fails to arrive at
relevant information. When retrieving unlikely associations or less common meanings of
ambiguous words (e.g., match in the context of a sport game vs. to light a fire), one must bypass
dominant meanings and associations and engage in a more deliberate and effortful search
through semantic memory. On the other hand, during the retrieval process, multiple aspects of
semantic information may be activated, and semantic selection processes direct activation
toward context-appropriate information while inhibiting dominant but irrelevant aspects of
knowledge. Importantly, these constructs are underpinned by distinct brain regions: controlled
retrieval processes are managed by the left anterior IFG (BA 47/pars orbitalis), while semantic
selection is mediated by the left mid-IFG (BA 45/pars triangularis). Moreover, semantic
selection may be a domain-general process, as the left mid-IFG is activated not only by
semantic tasks but also by various cognitive tasks requiring interference resolution and
selection, such as working memory, task switching, and response selection (Badre & Wagner,
2005; Brass & von Cramon, 2004; Jiang & Kanwisher, 2003; Thompson-Schill et al., 2002).

27
This indicates that the left mid-IFG responds to different stimuli by resolving competition
among active representations across multiple cognitive domains, highlighting its more domain-
general role in cognitive control and selection processes.

1.4.4. More than meets the I(FG)

Recent work has established a broader semantic control network (SCN), consisting of
interconnected regions beyond the LIFG, including the pMTG, pre-supplementary motor area
(pre-SMA) and dmPFC (Jackson, 2021). Studies using TMS have demonstrated that retrieval
mechanisms are equivalently disrupted when either the IFG or pMTG are targeted, specifically
in tasks with high semantic control demands (rather than low control or non-semantic tasks;
Whitney et al., 2011). Stimulation of the LIFG also magnifies task-related effects in the pMTG
and pre-SMA, highlighting the compensatory mechanisms within the extended semantic
control network (Hallam et al., 2016). Further, structural covariance between key semantic
control sites, reflecting greater cross-talk between regions, is related to individual differences
in semantic control abilities (Wang et al., 2018).
The literature proposes a distinct functional role for the pMTG in controlled, rather than
automatic, retrieval. For example, TMS studies have demonstrated that stimulation of the
pMTG disrupts judgments about weak thematic associations and identity matching tasks
requiring greater control (Davey et al., 2015). Consistent with this, Davey et al. (2016) found
activation of the left pMTG in the controlled retrieval of global semantic associations that
require the intentional linking of weak associations. Additionally, the pMTG also showed
resting-state and functional connectivity to the ventral IFG (BA 47) that has been implicated
in controlled retrieval processes. These findings suggest that a distributed semantic control
network, including the IFG and pMTG, is engaged when control is needed to shape retrieval
to suit the current context. In contrast, semantic processing guided by the automatic spread of
activation, such as when retrieving strongly encoded associations, has been associated with the
AG. Davey et al. (2016) also reported that the AG showed resting-state connectivity to regions
in the ATL, mPFC and PCC, areas which are a part of the Default Mode Network (DMN), a
large-scale brain network that has been linked to the automatic aspects of semantic processing.
This connectivity pattern underscores the AG’s role in supporting the effortless retrieval of
well-established semantic knowledge. Beyond controlled retrieval, the pMTG has also been
implicated in understanding thematic associations, rather than taxonomic categorical
information, and is specifically engaged during weak thematic associations (Teige et al., 2019).
This consistent engagement in representing thematic associations suggests that the pMTG may

28
facilitate controlled retrieval processes by dynamically updating contextual context in order to
retrieve situationally relevant information (Jefferies et al., 2020). In sum, the evidence
presented above highlights the differential roles of the pMTG and AG in retrieval, with the
pMTG supporting goal-directed retrieval of contextually relevant information, while the AG
facilitates more automatic integration of established knowledge.

1.4.5. The interaction between representation and control

The Controlled Semantic Cognition framework has received empirical support from studies
using functional connectivity analyses to demonstrate that semantic cognition involves a
dynamic interplay between the hub-and-spoke representational system and the semantic control
network (e.g., Chiou et al., 2018). Chiou and colleagues (2018) found that the IFG, a key node
of this system, increased its connectivity with spoke regions when participants performed tasks
that were less practiced and atypical, thereby requiring the intentional integration of
information or the examination of semantic attributes (e.g., pairing unrelated objects based on
colour). Building on this, Zhang et al. (2021) investigated controlled retrieval mechanisms
when participants are cued to search for a particular kind of information. The IFG was activated
during both goal-directed (top-down) and stimuli-driven (bottom-up) controlled retrieval, i.e.,
when participants are explicitly told what to retrieve in advance, and when they identified
associations between weakly linked concepts. Crucially, the study demonstrated that when
participants are asked to judge the semantic relatedness between words, the IFG classified goal-
information related to semantic relationships during the presentation of the first cue word. This
suggests that the IFG prepares and modulates the activation of relevant spoke regions, such as
the visual cortex, ensuring efficient semantic retrieval based on task goals. Further, goal-
information influenced activation in the visual cortex, suggesting that this task knowledge was
‘gating’ the flow of information. The IFG thus biases retrieval of relevant semantic attributes
from ‘spoke’ regions, dynamically controlling semantic processing by influencing early
information flow and retrieval strategies.

29
Figure 1: Masks of Networks of Interest. (Blue = Default Mode Network; Green = Semantic
Control Network; Red = Multiple Demand Network)

Beyond the SCN, semantic cognition also relies on domain-general executive control
mechanisms, underpinned by the Multiple Demand Network (MDN), which includes
frontoparietal regions in the dorsolateral PFC, precentral gyrus, pre- and supplementary motor
area, anterior and mid cingulate cortices and intraparietal sulcus (Duncan, 2010; Fedorenko et
al., 2013). This network supports goal-directed behaviour, such as problem-solving, planning,
and the ability to switch between tasks or strategies, through top-down executive control.
However, the retrieval of specific associations from semantic memory relies on more
specialized regions. Indeed, while there is some overlap between the SCN and the MDN, key
areas such as the LIFG and pMTG seem to be specialized for semantic control, whereas the
MDN is restricted to dorsal frontal and parietal areas (Jackson, 2021). Moreover, while the
SCN is predominantly left-lateralized, the MDN is a more bilateral network (Gonzalez Alam
et al., 2019). Topographically, key semantic control regions – the IFG and pMTG - are
positioned between the DMN and MDN, demonstrating a pattern of structural and functional
connectivity which influences their functional profile (Chiou et al., 2023a; Davey et al., 2016;
Margulies et al., 2016; Wang et al., 2020). Specifically, these regions showed heightened
responses for difficult semantic tasks and easier visuospatial tasks, indicating a specific tuning
for semantic difficulty that reflects a blend of the specialization of the MDN and DMN in
cognitive control and conceptual representation respectively (Chiou et al., 2023). Experimental
evidence for this distinction was presented by Gao et al. (2021), who manipulated the difficulty
in tasks of semantic judgment and verbal working memory to delineate common and distinct
neural mechanisms supporting control processes in these domains. While both tasks activated
MDN regions, including the dmPFC and pSMA, more difficult semantic tasks preferentially
activated the semantic control network, particularly the LIFG. Moreover, while the MDN
showed shared patterns of activation for both semantic and non-semantic demands, the SCN
did not, further bolstering the view that the SCN is specialized for semantic control.

30
1.5. Chapter Summary
In this chapter, I have outlined the current neurocognitive model of Controlled Semantic
Cognition, which proposes that effective semantic cognition depends on the interaction
between distinct systems of representation and control. Under this framework, knowledge
about the world is stored in a distributed manner across various sensory-motor brain regions,
each contributing modality specific information; while a central amodal hub, located in the
anterior temporal lobes, integrates this information to form coherent, multimodal semantic
representations. The semantic control system, underpinned by a widespread network of brain
regions including the IFG and pMTG, enables the flexible use of this knowledge by shaping
retrieval to meet current contextual demands. Recent work has indicated that the mechanisms
of semantic control extend beyond traditional semantic processing to other complex cognitive
functions, such as social processing (Diveica et al., 2021). Here, semantic control helps
modulate the retrieval and selection of socially relevant knowledge to ensure appropriate and
context-sensitive information. As I will explore in the next chapter, this idea can be applied to
the realm of creative thinking as well: control mechanisms can help facilitate the generation of
new ideas by allowing for the selective access and integration of atypical information.
Understanding how semantic control processes operates in these broader contexts is crucial, as
this can help provide deeper insight into the workings of the semantic system.

31
 Chapter 2

Creativity

2.1. Definitions
At the outset, it is important to clarify the level of creativity that we are interested in.
When thinking of ‘creativity’, the immediate association is with the extraordinary feats of
creative prowess demonstrated by eminent writers, scholars, and artists (referred to as ‘Big-C’
creativity; Kaufman & Beghetto, 2009). But creative thinking is at the core of our everyday
lives: when you do not have an ingredient and you replace it with something else; thinking of
an alternative route to your destination; or even rearranging the furniture in your living room
(called ‘little-c’ creativity; Kaufman & Beghetto, 2009). These everyday creative actions are
ubiquitous, from the workplace to our social relationships, enriching the experiences of
ourselves and our immediate others. This ‘little-c’ creativity has been extensively measured
using questionnaire-based methods, i.e., assessing self-reported creative beliefs and behaviours
(Diedrich et al., 2018; Dollinger, 2003; Kaufman, 2012), as well as personality traits (Oleynick
et al., 2017) and thinking styles (Fürst & Grin, 2018); and using task-based measures (e.g.,
divergent thinking tasks, insight problem solving; Cortes et al., 2019), where the focus is on
the creative product.

2.1.1. Product vs Process

In the literature, the standard definition of creativity has two criteria: novelty and
effectiveness (Barron, 1955; Runco & Jaeger, 2012; Stein, 1953). Undeniably, if an idea is not
original, it fails to meet the basic requirements for creativity; but originality alone is
insufficient: a truly random idea could be extremely unique or unusual but may not have any
actual use. The ‘effectiveness’ condition has been labelled as ‘useful or tenable’ (Stein, 1953),
‘task appropriate’ (Kaufman, 2016) or ‘valuable’ (Hennessey & Amabile, 2010). While
several other dimensions have been suggested (e.g., ‘surprise’; Simonton, 2012), the literature
has consistently used novelty and usefulness to evaluate creative output. While this standard
definition has been extremely beneficial for assessing the quality of creative output, it is
necessary to make a distinction between creativity as a product and as a process (Rhodes,
1961). The word has dual usage, referring to ‘creative-ness’ as an attribute of a person or
product (e.g., ‘What a creative ending to that story!’); and ‘creativity’ as a process constituted
of actions. An overreliance on the product-based definition has limited our understanding of

32
the cognitive mechanisms behind creativity. Green et al., (2023) suggest that a process-based
definition with three core elements: (1) creativity requires internally generated attention, i.e.,
representations from memory; (2) creative processes are guided by goal-related constraints; (3)
the end goal of the creative process is to generate a new representation that is not already held
in memory. Conceptualising creativity in this manner aligns with neurocognitive findings and
models, providing a more comprehensive framework for understanding creative thought.
Further, it highlights the wide applicability of this work, e.g., for academic success (Gajda et
al., 2017) and in the workplace (Reiter-Palmon et al., 2012). This thesis contributes to this
understanding by investigating both the processes underlying creative thought and the resulting
creative products.

2.1.2. Divergent and Convergent Thinking

Guilford’s Structure of Intellect Model (1950) put forth two types of creative operations
that have been fundamental to the study of creativity: divergent and convergent thinking.
Divergent thinking involves the generation of multiple original responses, whereas convergent
thinking requires a focused search for one correct solution. Divergent thinking is the most
common indicator of creative potential (Runco & Acar, 2012) usually measured using open-
ended tasks such as the Alternate Uses Task (AUT) in which participants are asked to generate
multiple uncommon and effective uses for a common object. Performance on the task is scored
primarily through metrics of fluency (i.e., number of responses), uncommonness of responses
(i.e., the unusualness of the response based on the sample), flexibility (i.e., the number of
unique uses), and subjective ratings (i.e., ratings of creative quality). On the other hand,
convergent thinking is measured using tasks with multiple cue constraints, such as the Remote
Associates Task (RAT; Mednick, 1962), wherein participants are given three cue words and
asked to find a target word that relates to all three cues. The RAT is often criticized as not being
a ‘creativity’ task, but this is based on analysis of the product (which is highly constrained),
not the process. Although RAT problems have a single correct solution, the process of reaching
the solution involves thinking about cue words in novel ways to meet a specific goal, hence
fitting the process definition of creativity.
Guilford (1950, 1956) initially framed divergent and convergent thinking as mental
operations that contributed to intelligence, but also emphasised divergent thinking as a key
component of creativity, which he considered to be a separable dimension from intelligence.
Wallach & Kogan, (1965) made a stronger distinction between the two, portraying divergence
as associative processing, and convergence as executive processing. However, more recent

33
work has suggested that neither of the AUT nor the RAT is purely divergent or convergent –
rather, successfully performing these tasks requires both processes (Benedek & Jauk, 2018;
Cortes et al., 2019; Dygert & Jarosz, 2020). For example, on the AUT, participants must ensure
that the generated use is not only novel, but useful, indicating that they must engage in a
convergent process to evaluate and select from competing ideas. Similarly, to successfully
solve the RAT, one must generate possible solution words (a divergent process), as well as
evaluate and select ideas in relation to the cue words (a convergent process). Indeed, while
these have been portrayed as dissociable constructs, more recent formulations suggest that they
are, in fact, intertwined during the creative process – with divergence broadening the search
space, and convergence identifying the best aspects of that space (Cortes et al., 2019). In line
with this, factor analytic evidence suggests that these tasks rely on distinct and shared cognitive
processes (Dygert & Jarosz, 2020). In summary, while divergent and convergent thinking are
often treated as distinct processes, they are inextricably linked during creativity. This
intertwined relationship between divergent and convergent thinking highlights the need to
examine specific cognitive processes, such as those related to semantic control. By exploring
these mechanisms, this thesis aims to provide a deeper understanding of the cognitive processes
underlying creativity.

2.2. Situating Semantic Memory within Creativity Theories

Current theories of creativity can be classified into two primary schools: associative
and executive accounts. Associative accounts propose that individual differences in semantic
memory structure and associative abilities facilitate creative thinking (Beaty & Kenett, 2023;
Mednick, 1962); while executive accounts argue that creativity emerges from individual
differences in top-down control processes, including fluid intelligence (i.e., reasoning ability)
and other executive functions (Beaty, Silvia, et al., 2014; Beaty & Silvia, 2012, 2013; Benedek,
Franz, et al., 2012; Silvia & Beaty, 2012). Nevertheless, converging evidence from cognitive,
computational, and neuroscientific research, reviewed in the following sections, points towards
a dual-process account, suggesting that creativity emerges from a complex interplay between
associative and executive processes (Kleinmintz et al., 2019; Sowden et al., 2015).
Semantic memory is fundamentally important to creative thinking (Abraham, 2014;
Abraham & Bubic, 2015). This is evident in both accounts of the creative process – in
associative accounts, semantic memory is directly important due to its contents and structural
properties; while in executive accounts, it provides the raw material that higher-order cognitive
processes manipulate. The relevance of semantic memory to creativity is illustrated in lesion

34
study by Paulin et al. (2020): compared to healthy controls, semantic dementia patients
produced much fewer and less original responses to the AUT, highlighting how the degradation
of the semantic system diminishes the knowledge base necessary for creative thinking.
Moreover, divergent thinking is related to crystallized intelligence (i.e., knowledge of facts,
vocabulary and general information accumulated from education and experience), indicating
that individuals draw on the breadth and depth of what they know to form new ideas (Batey et
al., 2009; Cho et al., 2010). This relationship suggests that the breadth and depth of an
individual’s stored knowledge play a role in their ability to generate novel ideas. In line with
this, a recent meta-analysis demonstrated that semantic memory, compared to episodic
memory, working memory and short-term memory, was reliably associated with divergent and
convergent thinking (r=.20; Gerver et al., 2023). However, it is important to note that many of
the semantic memory tasks used in this study involved strategic retrieval processes,
underscoring the importance of controlled semantic cognition for creative thought.

2.3. Associative Theories

Association (i.e., the linking together of concepts) is fundamental to the creative
process. According to Mednick (1962), creativity refers to the building of associative links
between concepts in alignment with external goal constraints, where more distant conceptual
combinations result in more creative outcomes. Under his framework, concepts are organized
in associative hierarchies (i.e. a set of associations arranged by the strength of their
relationships) and individual differences in the capacity for association are at the heart of
creative ability. Under this conceptualisation, as in the classic spreading activation models in
the semantics literature (Collins & Loftus, 1975), when a concept is activated, there is an
automatic spread of activation to related concepts through these associative links. Mednick
(1962) further proposed that these individual differences in associative abilities arise from the
structure of the semantic networks. Less creative individuals have ‘steep’ associative
hierarchies, characterised by a predominance of stereotypical connections of strong associative
strengths (e.g., bread-butter), but fewer remote connections with weak associative strength
(e.g., bread-money). Conversely, creative individuals are more able to access and combine
remote associations, due to their ‘flat’ associative hierarchies, wherein a given concept has
connections of similar strengths to both common and distant associates (e.g., bread would be
similarly linked to both butter and money). Thus, the flatness allows creative individuals to
avoid dominant associations in favour of more original associations.

35
2.3.1. Evidence from Computational Network Studies
Network science, based on mathematical graph theory, has validated and extended
Mednick’s original theory by offering quantitative methods to investigate the structure of
semantic memory. In these computational network-based studies, semantic knowledge is
represented as a network of interconnected concepts (Hills & Kenett, 2022), composed of nodes
(i.e., concepts) and edges (i.e., the association between concepts). In line with Mednick’s
theory, these approaches have established that the structure of the semantic network influences
creative thinking (Benedek et al., 2017; Kenett et al., 2014; Kenett & Austerweil, 2016; Luchini
& Beaty, 2023). Additionally, this work has identified key network properties of high creative
individuals that may allow for the more efficient retrieval and combination of distant concepts:
flexible and densely connected networks, with short average path lengths, and low modularity
(Benedek et al., 2017; Kenett et al., 2014; Kenett & Austerweil, 2016). Densely inter-connected
networks make it easier for individuals to access a range of ideas quickly; while shorter path
lengths indicate that concepts can be connected by traversing fewer intermediate nodes,
allowing individuals to jump from one idea to a distant associate. Moreover, low network
modularity indicates that the network is not easily divided into subcommunities, confirming
Mednick’s (1962) proposal that creativity is underpinned by flatter, rather than steeper,
hierarchies. Finally, the semantic networks of more creative individuals are flexible, as they
exhibit greater resilience to simulated targeted attacks, as evidenced by their slower network
breakdown (Faust & Kenett, 2014). This resilience is due to the network properties outlined
above that allow for the more efficient flow of information by ensuring widespread, non-
isolated network connectivity, and areas of local redundancy that allow for multiple pathways
for information flow. Taken together, these findings indicate that creative individuals possess
semantic network structures that are amenable to the more efficient retrieval of remote
information. However, while denser semantic networks can benefit idea quantity by providing
a vast array of associations that facilitate the spread of activation; this may come at the cost of
increased interference, leading to poorer idea quality (Beaty et al., 2023).

2.3.2. Individual Differences Approach

While the structure of the semantic network provides the foundation for the efficient
spread of information, it is the individual's associative abilities that drive the process, enabling
them to leverage this efficiency in pursuit of their creative goals (He et al., 2021). The
individual differences literature supports the notion that creative people have stronger
associative abilities and can generate a wider and more diverse array of unusual responses than

36
less creative people (e.g., Beaty et al., 2021; Beaty & Silvia, 2012; Benedek, Könen, et al.,
2012; Benedek & Neubauer, 2013; Marron et al., 2018). More creative individuals estimate
unrelated word pairs to be closer together in semantic space (Rossmann & Fink, 2010); can
utilise both close and remote primes to make associations (Gruszka & Necka, 2002); and are
faster at judging whether two concepts are related (Vartanian, 2009). Further, they tend to
traverse more of their semantic space when performing chained-free association tasks used to
capture dynamic memory retrieval processes (Gray et al., 2019). Taken together, these studies
strongly suggest that the ability to make quick and more distant associations is critical to
creative thinking.
However, while Mednick emphasized the critical role of semantic memory structure in
creativity, his theory does not account for the critical role of top-down executive control
processes. In a direct empirical test of Mednick’s main hypothesis, Benedek and Neubauer
(2013) found no difference between the associative hierarchies of low and high creative
individuals, implying similarities in the general organisation of associative memory between
these groups (e.g., when cued with ‘street’, both are likely to respond ‘car’). This finding
suggests that shared environments and experiences may establish common associations for
both groups. The key difference between groups was that more creative people generated more
uncommon responses, but this was driven by associative fluency (i.e., they were quicker at
generating associations) which allowed them to reach more distant concepts faster. These
findings imply that creativity may involve executive abilities that allow for efficient retrieval
and manipulation of semantic memory.
Moreover, when searching for creative uses for common objects, responses tend to
display a serial order effect wherein the creative quality of ideas increases over time (Beaty &
Silvia, 2012). This alludes to the work of associative processes, with initial responses drawing
from readily accessible associations in semantic memory. As these typical ideas are exhausted,
individuals dig deeper within their semantic networks, leading to more unique associations.
However, the serial order effect diminishes as fluid intelligence increases, indicating that
people with very high fluid intelligence start off with better ideas and the quality of their ideas
increase through the task. Indeed, individual differences in fluid intelligence and broad retrieval
ability predict creativity, even after controlling for associative ability, further supporting the
idea that both associative and executive abilities independently contribute to creativity. This
pattern is seen in convergent thinking tasks as well: solving RAT problems successfully is
driven by the ability to avoid typical, high frequency associates, and to correctly assess the
feasibility of the response options (Gupta et al., 2012). This strongly points toward a role of

37
executive functions in divergent thinking, particularly in overriding dominant and closely
related ideas in favour of more remote associations.

2.4. Executive Accounts of Creative Cognition

The controlled-attention theory positions creativity as emerging from individual
differences in top-down processes of attention and cognitive control (Beaty, Benedek, et al.,
2014, 2014; Beaty & Silvia, 2012, 2012, 2013; Benedek, Franz, et al., 2012). According to
these accounts, control processes work to constrain the associative process, in line with task or
contextual requirements, through the use of strategies, rules, controlled retrieval of specific
information, and evaluative processes. In line with this, multiple studies have established a
relationship between creative thinking and various domain-general executive abilities (i.e.,
cognitive functions that are crucial for planning, decision-making, and adapting behaviour in
response to changing goals or environmental demands) and facets of intelligence. In the
following sections, I will first review evidence for the relationship between creative thinking
and domain-general executive functions, as well as intelligence. Then, I will explore the
connections between these accounts of creativity and the construct of semantic control
introduced in the previous chapter, highlighting its crucial role in managing and directing
creative thought.

2.4.1. Executive Functions

Evidence for the role of executive control in creativity comes from individual differences
research, which examines the contributions of cognitive control abilities to divergent thinking
and other measures of creative potential, such as creative metaphor generation (Beaty & Silvia,
2013; Silvia & Beaty, 2012); humour production (Kellner & Benedek, 2017); and insight
problem solving (Lee & Therriault, 2013). Of specific interest are executive functions, domain-
general mechanisms that regulate cognitive processing in a goal-directed manner (e.g.,
Diamond, 2013; Friedman et al., 2008). Miyake et al. (2000) identified three core executive
functions: updating (monitoring/manipulation of working memory contents); shifting (flexible
switching between tasks); and inhibition (intentional avoidance of prepotent responses). These
executive functions are highly correlated but independent constructs (Friedman et al., 2008),
and are crucial for higher order cognitive abilities, such as fluid intelligence (Diamond, 2013).
A meta-review of 57 studies indicates an inconsistent relationship between creativity and
these core executive functions (Palmiero et al., 2022). Most reliably, the review found that
divergent thinking, especially creative fluency (i.e., the ability to generate multiple creative
ideas), is related to updating abilities (Batey & Furnham, 2006; De Dreu et al., 2012; Nusbaum

38
& Silvia, 2011). This indicates that the ability to maintain, monitor and update task-relevant
information facilitates the greater production of ideas, but not necessarily the quality of those
ideas. Next, divergent thinking is variably associated with inhibition, with studies suggesting
that inhibition can be both helpful (Benedek, Franz, et al., 2012; Zabelina et al., 2012) and
harmful (Cheng et al., 2016; Radel et al., 2015) to creativity. Thus, divergent thinking may
benefit from flexible inhibitory control that can be engaged to varying extents according to task
requirements. For example, inhibition of stereotypical information is critical to avoid irrelevant
detours, but sometimes, less salient information may be the key to the construction of original
links between concepts (Vartanian, 2009). Lastly, while there is a strong theoretical case for the
association between divergent thinking and shifting (i.e., ability to switch between different
types of information) the few studies that have examined this relationship have found
inconsistent results (Benedek, Jauk, et al., 2014; Zabelina et al., 2019). These inconsistencies
in the role of executive functions in creativity arise from a variety of methodological factors,
such as visual/verbal modality, scoring of divergent thinking and tasks used. For example,
while Benedek et al. (2014) found that updating and inhibition both predicted ratings of creative
quality, only updating was related to fluency and uniqueness in a study by Zabelina et al.
(2019). This discrepancy may be due to the number and type of tasks used to measure the
executive functions and the different scoring of divergent thinking in the two studies.

2.4.2. Fluid Intelligence

The relationship between intelligence and creativity has been controversial. Early
positions suggested that they are distinct, quite weakly related, factors (Batey & Furnham,
2006; Kaufman & Plucker, 2011; Kim, 2005). But more recent meta-analytic evidence suggests
a more meaningful relationship between intelligence and creativity (r=.25 to .37; Gerwig et al.,
2021), after accounting for moderating factors such as scoring methods, choice of
tasks/instructions, and time on task. This evidence suggests that creativity and intelligence are
more interconnected than previously thought, particularly when considering the different facets
of intelligence, such as fluid and crystallized intelligence.
Divergent thinking is reliably associated with both fluid and crystallized intelligence,
consistent with the idea that both the contents of our conceptual system and the ability to
regulate the use of the knowledge are necessary for creative thinking (Gerwig et al., 2021).
Fluid intelligence reflects reasoning abilities (i.e., the capacity to solve problems independent
of prior knowledge), that enable individuals adapt to new situations (Cattell, 1971; Schneider

39
& McGrew, 2018); while crystallized intelligence represents the knowledge and skills
accumulated from experience and education.
But, what is the functional role of intelligence in creativity? Nusbaum and Silvia (2011)
found that when participants are explicitly encouraged to use a strategy, fluid intelligence is
more predictive of divergent thinking, indicating that people who are higher in fluid
intelligence can benefit more from implementation of strategies that are useful but cognitively
costly. Further, fluid intelligence is more predictive of creativity when people are explicitly
told to focus on the creative quality of their response, indicating that, here again, intelligence
facilitates top-down strategy use (Nusbaum et al., 2014). Finally, fluid intelligence is critical
when knowledge about an object is sparse, indicating that it may facilitate the deliberate
combination of remote associates; and when there is semantic interference in densely
connected networks, illustrating the role of top-down control in the inhibition of dominant ideas
(Beaty et al., 2023).
Strategy use is common across divergent and convergent thinking tasks. Studies
adopting a ‘think aloud’ procedure (i.e., asking participants to verbalise their inner thoughts)
demonstrate the widespread usage of strategies when performing the AUT (Gilhooly et al.,
2007). Notably, however, strategies like simple long-term memory retrieval or self-cueing (i.e.,
repeating the name/features of the object) did not facilitate creative responses; instead, creative
responses were more likely when more abstract strategies were used (e.g., breaking apart the
object). This style of strategy use suggests that, since divergent thinking tasks present open-
ended and ill-defined problems, individuals may define more attainable sub-goals through their
choice of strategy. Identifying these sub-goals can activate a series of steps that may lead to the
answer, with individuals generating solutions to reach each sub-goal and evaluating these
according to task constraints. This strategy use is also seen when solving RAT problems, with
participants relying on a dominant cue to guide their search through semantic memory, or
constraining their associations based on their previous responses (Smith et al., 2013). Thus, to
be creative, one must select and apply a useful strategy, the implementation of which may be
facilitated through enhanced fluid intelligence abilities.

2.4.3. A note on the benefits of reduced control

Top-down attentional control, a core component of executive function, is also related to
creativity (Beaty, Silvia, et al., 2014; Benedek & Jauk, 2018). The ability to exercise executive
attentional control mediates the relationship between fluid intelligence and divergent thinking
(Frith et al., 2020), implying that focused attention benefits divergent thinking by allowing for

40
the maintenance of goal directed internal processing (Benedek, Franz, et al., 2012; Benedek,
Jauk, et al., 2014) that guides memory search and facilitates remote conceptual combinations
(De Dreu et al., 2012; Gilhooly et al., 2007; Zabelina, 2018). However, very stringent idea
evaluation may negatively affect idea generation, resulting in the production of fewer creative
ideas (Ivancovsky et al., 2019). At times, ‘leaky’ bottom-up attention is also necessary, as this
diffused attention allows a wider range of stimuli to enter awareness, reducing latent inhibition
and stimulating the generation of creative ideas. Importantly, creative individuals have flexible
attentional control, easily switching between focused and broad modes of attention (Gabora,
2010; Vartanian, 2009; Zabelina & Robinson, 2010). In line with this, some neuropsychological
work suggests that less prefrontal engagement may be beneficial for creativity. In these cases,
damage to left temporoparietal and inferior frontal regions leads to an increase in creative
abilities by releasing inhibitory control over the generative system (e.g., Abraham et al., 2012;
Mayseless et al., 2014; Shamay-Tsoory et al., 2011). This is also supported by TMS studies
inhibiting left IFG activity (Chrysikou, 2019; Ivancovsky et al., 2019; Kleinmintz et al., 2018a).
For example, inhibiting the left IFG using continuous theta-burst stimulation (cTBS) leads to
the generation of more original ideas; but, suppression of IFG activity during idea evaluation
led to a reduction in the perceived deviance of ideas (Kleinmintz et al., 2018a). Taken together,
these results suggest a complex interplay between attention and its effects on creativity: at
times, excessive evaluation might hinder creativity by inhibiting generative processes, but
relaxing these inhibitory tendencies may allow us to gain fresh insights from the world around
us.

2.4.4. Free Association vs Goal Directed Association

At this juncture, it is important to draw a key distinction between free association and
goal-directed association, both of which have been studied in the creativity literature. Free
association tasks involve generating as many associations as possible to a cue word, without
any specific constraints, typically resulting in the automatic retrieval of information from
memory (Fradkin & Eldar, 2023; Richie et al., 2023). In contrast, goal-directed association
requires the generation of responses that align with task requirements, meaning that people
must extract specific information from memory e.g., listing animal names (Silvia et al., 2013;
Todd & Hills, 2020). Under the Cattell-Horn-Carroll model, this goal-directed association has
been operationalised as ‘broad retrieval ability’ (Carroll, 1993; Schneider & McGrew, 2018),
encompassing a range of narrow fluency-related abilities that reflect the ease and flexibility
with which individuals can generate ideas. Broad retrieval has been found to be a predictor of

41
creativity (Gerver et al., 2023; Gerwig et al., 2021; Miroshnik et al., 2023), but this relationship
is mediated by general processing speed (Forthmann, Jendryczko, et al., 2019; Miroshnik et
al., 2023), indicating that faster cognitive processing enhances the efficacy of retrieval.
However, tasks tapping into broad retrieval ability tend to emphasize the retrieval of multiple
ideas, emphasising the breadth of idea generation, often at the expense of depth and specificity.
Broad retrieval ability is commonly assessed through various fluency tests, which
generally require participants to quickly generate as many responses as possible in response to
a cue word (Forthmann, Jendryczko, et al., 2019; Silvia et al., 2013). These tasks include
semantic fluency tasks, where participants are prompted to generate category examples ('jobs’,
Gilhooly et al., 2007; Unsworth et al., 2011); phonemic fluency tasks which require participants
to generate words with a given wordstem (e.g. ‘ab’); and associational fluency tasks which
require participants to generate synonyms for a word. While both language and executive skills
are thought to be determinants of verbal fluency (Henry & Crawford, 2004), more recent
evidence suggests that both phonemic and category fluency are primarily driven by language-
related capacities such as vocabulary (Whiteside et al., 2016). Moreover, different cognitive
processes may underlie various fluency tasks. For example, category fluency relies on free
association and spontaneous semantic processing; while phonemic fluency requires executive
control to supress dominant activation, thereby engaging more deliberate search processes
(Marko & Riečanský, 2023). Importantly, this highlights differences between fluency tasks that
prioritize the quantity of ideas generated, and controlled retrieval tasks that call for the retrieval
of specific, often weakly-related targets.
Additionally, broad retrieval and divergent thinking tasks have some overlap in their
content, instructions, and coding of responses, particularly with regard to metrics of response
fluency (Miroshnik et al., 2023). While both tasks involve selective retrieval mechanisms,
verbal fluency tasks require retrieval of typical exemplars, and divergent thinking tasks depend
on the extraction of atypical associations. Therefore, divergent thinking tasks may place higher
demands on selective retrieval than verbal fluency tasks, as the generation of novel ideas
necessitates the suppression of more common items that might be activated during search. This
suggests that effective creative thinking likely requires both broad retrieval for generating
numerous ideas, and controlled processes for the goal-directed retrieval of specific information.
In sum, I have argued that there is reason to suggest that while broad retrieval ability is related
to divergent thinking, it may not fully capture the goal-directed processes essential for create
ideation. Instead, specific semantic control mechanisms may be more critical to creative
cognition, strategically guiding the retrieval and combination of ideas.

42
2.4.5. Semantic Control

Despite the clear emphasis on semantic knowledge as a foundation for creative thought,
the role of semantic control mechanisms in creative processing remains underspecified. In the
previous chapter, I elaborated on the nature of these semantic-specific control processes,
emphasizing how they are engaged in situations where there is a high demand for retrieving,
regulating and selecting relevant semantic information. Building on this, there are several
compelling reasons why semantic control may be crucial for creativity. We possess vast
amounts of knowledge, including information about the various features and attributes of a
concept, and how it relates to other concepts and contexts. However, when trying to generate
new ideas, only a small portion of this information may be relevant, making it necessary to
inhibit dominant associations and strategically access atypical information when searching
through memory.
Let us consider the example of thinking of novel uses for a belt. One may want to first
inhibit stereotypical associations of belts as items of clothing to consider more unconventional
ideas. Thus, when the passive spread of semantic activation does not result in a desirable
outcome (here, something that is novel and useful, in line with task constraints), controlled
retrieval abilities can contribute to creative thinking by maintaining retrieval goals and
facilitating an effortful search for less dominant semantic information. This could involve
considering more niche semantic features of the belt – such as its flexibility, length, or texture
– in order to create new uses. For instance, focusing on the flatness of a belt might inspire using
it as a plate for sushi, or considering its sturdiness could lead to the idea of making a makeshift
bookshelf. In turn, this generative process triggers the retrieval of multiple competing pieces
of information, necessitating semantic selection processes to evaluate and select from these
potential ideas in line with task demands. For example, the association of the belt with food
might trigger a new idea of using a belt as a cake stand, but this could be evaluated as
insufficiently novel after producing the similar sushi use. This may involve not only selecting
which semantic features of a concept to focus on but also reinterpreting these features in new
and unconventional ways. There has been considerable neuroimaging work implicating key
regions of the semantic control network in creative tasks, as I will review in the following
sections. However, to date, there has been scarce behavioural work examining the role of
semantic control processes in creative thinking, and more specifically, in assessing the
individual contributions of controlled retrieval and semantic selection.

43
2.5. Dual Process Models
Taken together, the above literature can be synthesized as follows: creativity emerges from
a complex interaction between bottom-up associative processes, acting on semantic memory,
and top-down processes that enable us to flexibly extract and use relevant representations
(Beaty, Silvia, et al., 2014; Zhuang et al., 2021). This view parallels Kahneman’s two systems
approach, wherein two modes of thinking, Type 1 and Type 2, have commonly been proposed
(Evans, 2008). Type 1 processes are rapid, unconscious, automatic and effortless; while Type
2 processes are slow, conscious, controlled and effortful (Kahneman, 2011). Goal-directed
thought requires cognitive control for maintaining task relevant information, usage of
strategies, progress monitoring and outcome evaluation. Indeed, Barr et al. (2015) provided
empirical support for the relevance of Type 1 and Type 2 processing to creativity. Their work
demonstrated that while initial creative insights are linked to Type 1 processes, the development
and elaboration of these ideas, particularly in complex tasks, rely heavily on Type 2 processes.
For instance, when participants engaged in tasks requiring the combination of distant
connections, effective performance necessitated the overriding of intuitive responses in favour
of more deliberate reasoning. This aligns with the notion that creativity benefits from a balance
between spontaneous and controlled processing.
Similarly, many variations of this dual-process model have been proposed in the creativity
literature, such as the Blind Variation and Selective Retention Model (Campbell, 1960); the
Glenpore Model (Finke et al., 1992); Honing Theory (Gabora, 2010); the Dual Pathway Model
(Nijstad et al., 2010); and the two-fold model of creativity (Kleinmintz et al., 2019). Together,
these models suggest that creativity is underpinned by a generation phase, involving rapid,
automatic and associative processes that lead to the production of ideas (akin to Type 1
processing); and a slow, effortful and controlled evaluation phase, where ideas are tested and
refined to fit task criteria (akin to Type 2 processing). While all the models agree on an
interaction between generative and evaluative phases, the differences lie in the mechanisms of
the interaction: some models suggest distinct cyclic phases where one type of thinking is
dominant; while others suggest a more fluid interaction where both processes operate in
parallel. The ‘distance-dependent representation activation mode’ (DDRAM Hypothesis; Volle,
2018) extends this fluid view by suggesting that controlled processing occurs not only at the
evaluation stage, but also during the generation phase. More specifically, control processes
work to constrain the generation phase, through the inhibition of stereotypical ideas, or the use
of strategies like focusing on specific aspects/features. This is important, as it reflects findings
from the literature that suggest that goal directed associative processes are recruited when

44
control demands are high, or unusual. Supporting these theoretical models, neuroimaging
studies have revealed functional coupling of brain networks related to associative and executive
processes during creative tasks (Beaty et al., 2015, 2016; Beaty, Silvia, et al., 2014; Ellamil et
al., 2012; Green et al., 2015; Mayseless et al., 2015; Takeuchi et al., 2012). This neurobiological
perspective will be discussed in the following sections.

2.6. Neural basis of creative thinking

Whole-brain functional connectivity analyses suggest that divergent thinking is
underpinned by a widely distributed network, including regions in the default (posterior
cingulate cortex, precuneus and inferior parietal lobule) and the executive networks
(dorsolateral prefrontal cortex; Beaty et al., 2015). In creativity, evidence suggests that the
default network contributes to the generation of ideas, while the executive network constrains
the generative process in line with task requirements (Beaty et al., 2016).

2.6.1. Associative processes

Creativity has been consistently linked to the default mode network (DMN), which includes
regions in the medial prefrontal cortex, posterior cingulate cortex, and various temporoparietal
regions (Fox et al., 2005; Raichle & Snyder, 2007). The DMN is possibly best known as a set
of regions that show consistent task-related deactivation, illustrating the role of this network in
spontaneous internal mental processes, such as mind wandering. However, the DMN is also
recruited during deliberate goal-directed processes like episodic retrieval, future thinking or
mentalizing (Andrews-Hanna, 2012; Buckner, 2012; Spreng et al., 2009). Thus, the default
network is characterised by internally directed cognition that involves constructive processes
building on the contents of memory, rather than on immediate sensory input (Andrews‐Hanna
et al., 2014; Schacter et al., 2012; Smallwood & Schooler, 2015). Given this role of the DMN
in imaginative processes, we can assume that its engagement in creativity tasks also reflects
spontaneous and self-generated thought (Abraham, 2014; Beaty, Benedek, et al., 2014).
In line with this, numerous studies have established the importance of the default network
to divergent thinking (Ellamil et al., 2012; Gonen-Yaacovi et al., 2013; Howard-Jones et al.,
2005; Mayseless et al., 2015; Shofty et al., 2022; Takeuchi et al., 2012; Wei et al., 2014).
Moreover, the strength of resting-state connectivity between key DMN regions is related to
superior divergent thinking skills. This DMN activity may be key to the generation of creative
ideas (Ellamil et al., 2012) and has been specifically linked to associative thinking. For
example, Marron and colleageus (2018) asked participants to generate words in response to a
single cue, forming an associative chain where each word relates to the one before it. Core

45
regions of the DMN (including the mPFC, PCC, left TPJ and MTG) and the left IFG were
consistently activated during this free association task, compared to phonemic/semantic
fluency or episodic memory tasks. Taken together, these findings indicate that the DMN
support creativity by facilitating the unconstrained, associative search through memory.
However, the involvement of the IFG is worth highlighting – as mentioned above, this region
is an integral part of the semantic control network – as this implies that there is some degree of
goal-directedness involved in the production of free association. This will be discussed further
in the following sections.

2.6.2. Executive Processes

Executive functions rely on a frontoparietal control network, which is referred to by a range
of terms in the literature, including the ‘Executive Control Network’, ‘Fronto-Parietal Control
Network’ and ‘Multiple Demand Network’. Throughout this thesis, I will refer to these regions
as the Multiple Demand Network (MDN; Duncan, 2010; Fedorenko et al., 2013). This
executive network, underpinned by regions in the dorsolateral and ventrolateral prefrontal
cortices, dorsal and anterior cingulate cortex and anterior IPL, is engaged during experimental
tasks requiring externally-focused attention and cognitive control processes (Seeley et al.,
2007). These areas facilitate domain-general executive control and goal maintenance through
processes like working memory, inhibition, and flexibility (Duncan, 2010; Fedorenko et al.,
2013). The role of the executive network in creativity illustrates the involvement of cognitive
control mechanisms including working memory, inhibition, and flexibility (Beaty, Silvia, et al.,
2014; Benedek, Jauk, et al., 2014; Kleinmintz et al., 2019). A recent ALE meta-analysis
highlighted the importance of executive control to divergent thinking across a range of tasks,
including the AUT, creative verb and metaphor generation, and creative visualisation (Cogdell‐
Brooke et al., 2020): it implicated regions of the MDN, including the dorsolateral prefrontal
cortex and superior frontal gyrus, in divergent thinking. This is consistent with behavioural
work evidencing the role of these top-down processes in creative thought (Beaty & Silvia,
2012; Silvia et al., 2013).
More importantly, the Cogdell-Brooke et al. (2020) meta-analysis established the critical
role of the ventrolateral prefrontal cortex to divergent thinking. This region is also important
for convergent thinking, with lesions in the left rostrolateral prefrontal cortex linked to
impairments in combining remote semantic associations (Bendetowicz et al., 2018). When
examining the component parts of the dual process model, studies have established that IFG
demonstrates increased engagement during the evaluation of creative ideas (Ellamil et al.,

46
2012; Kleinmintz et al., 2018b; Kröger et al., 2012; Mayseless et al., 2014). For example,
Kröger et al. (2012) found that IFG activity was increased when participants assessed a
response as being novel and appropriate, rather than commonplace/inappropriate, suggesting
that the IFG may be engaged when individuals search for the weak link between concepts.
However, the IFG is a key part of the semantic control network and may be important for both
the goal-directed search for original ideas, as well as the evaluation of ideas in line with task
requirements. This will be discussed in the following sections.

2.6.3. Interplay between the default and executive networks

The default and executive networks usually display an antagonistic relationship during
a host of attentionally demanding cognitive tasks, with the default network deactivating and
the executive network increasing activation with task demands. However, these networks
exhibit cooperation during tasks requiring the top-down regulation of internally generated
information, such as maintaining an internal train of thought (Smallwood et al., 2012). Here,
the executive network flexibly couples with the default network to support self-generated
thought, allowing the integration of information across cognitive domains and facilitating
complex cognitive processes such as planning, decision making and problem-solving. This
cooperative coupling allows the executive network to leverage the rich associative information
stored in the default network, and is seen across numerous different processes, such as episodic
future thinking, mental time travel, and constructive day dreaming (Spreng et al., 2010).
Numerous studies have established that an increased functional connectivity between
default and executive networks contributes to creative thinking across a range of different tasks
(Beaty et al., 2014, 2016; Green et al., 2015; Mayseless et al., 2015; Takeuchi et al., 2012; Zhu
et al., 2017) as well as artistic domains (Ellamil et al., 2012; Liu et al., 2015; Pinho et al., 2015).
Moreover, stronger patterns of connectivity between these networks, both in resting-state and
task-based fMRI, is associated with individual differences in creativity (Beaty, Benedek, et al.,
2014; Beaty et al., 2015, 2016; Beaty, Kenett, et al., 2018; Green et al., 2015). These network
dynamics have been proposed to support cognitive processes that underlie creativity, such as
memory retrieval and inhibition. For example, when participants’ episodic retrieval
mechanisms are primed, they generate more original object uses, and exhibit greater default-
executive coupling, reflecting goal-directed retrieval from episodic memory (Madore et al.,
2019). Similarly, this coupling facilitates inhibition of pre-potent responses, increasing when
control demands are high. For example, increased default-executive coupling is demonstrate

47
when participants are faced with high ideational constraints, illustrating the use of more
inhibitory control to avoid dominant associations (Beaty et al., 2017).
Interestingly, while both younger and older adults exhibit functional default-executive
coupling, this pattern is exaggerated in the older group (Adnan, Beaty, Lam, et al., 2019). Older
adults present greater coupling between the default regions and right inferior frontal gyrus: this
region has been identified as important for inhibition, suggesting that older adults may need to
work harder to suppress over-learned prior knowledge in order to find new connections. These
results are in line with the Default-Executive framework (DECHA, Spreng & Turner, 2019),
which suggests that, as people age, there is a shift toward a reliance on prior semanticized
knowledge.

2.6.4. Semantic Control Network

Beyond the DMN-MDN coupling, creativity may also rely on contributions from the
semantic control network (SCN). In previous sections, I put forward a case for the role of
semantic control in creative thinking: producing original ideas does not solely rely on
associative processes, but also requires the effective navigation and restructuring of one’s
semantic memory. The spatial location of the semantic control network, between the default
and executive networks, may allow it to facilitate the generation of creative ideas – the SCN
can strategically direct activation towards less dominant associations between concepts. The
SCN overlaps with the MDN but only partially – the SCN includes areas in the left anterior
and ventral parts of the IFG, which are not part of the MDN (Gao et al., 2021; Wang et al.,
2020). This argument is supported by meta-analytic evidence implicating the IFG as a key
player in creative cognition, facilitating semantic control processes underlying the generation
of new ideas (Cogdell‐Brooke et al., 2020; Gonen-Yaacovi et al., 2013). However, few studies
have directly investigated the role of the semantic control network in creativity directly. The
available evidence on this question is presented hereafter.
Krieger-Redwood and colleagues (2023) instructed participants to generate unusual
links between two cues and indicate whether the link was generated from knowledge or
personal experience. The authors manipulated the semantic distance between the cues (i.e.,
close, e.g., confetti-helium or distant, e.g., melon-bookcase). Close trials were linked to
episodic retrieval mechanisms and elicited similar responses from the sample; in contrast,
distant trials engaged a strategy involving the controlled retrieval of semantic information and
elicited more varied responses. This was reflected at the neural level as well, as semantically
similar trials engaged the DMN, while distant trials engaged the IFG. This implies that the

48
DMN and SCN may facilitate associative thought through different memory systems: the DMN
may utilise episodic representations of past experiences to form familiar associations, and the
SCN regions may contribute to forming new conceptual links in atypical context where past
experiences cannot be relied on.
In line with this, Beaty and Kenett (2023) suggest that the neurocognitive architecture
underlying associative thought may involve three networks: the default network, to support the
generation of new ideas via associative thinking, the semantic control network, to identify the
most appropriate ideas, and the executive network for the evaluation and manipulation of ideas.
This incorporates both controlled and spontaneous processes in unconstrained free association.
Aligning with the notion that the IFG “prunes” the search for ideas in line with task demands,
Vartanian et al. (2018) found that the IFG exerts control over activation in the middle temporal
gyrus and inferior parietal lobe during divergent thinking. Similarly, this pattern is also seen
during convergent thinking: Becker et al. (2020) highlight the role of semantic control areas in
both the search and solution phase of the RAT. Key areas of the semantic control network,
including the left MTG and IFG were more engaged in the search phase; while a more bilateral
network was activated for the solution phase. Further, the results suggested that the IFG may
play a distinct role in these two phases: in the search phase, it may be crucial for inhibition of
associations and retrieval of possible solutions; while during solution, it may be involved in
competition resolution and selection processes. This distinction highlights the IFG's role in
both inhibiting irrelevant information and facilitating the retrieval of pertinent associations,
making it integral to the creative problem-solving process.
Most creativity tasks require one-word responses to stimuli, but more recent studies
have been employing longer form measures. For example, Rastelli et al., (2024) gave
participants a story generation task, varying the levels of semantic control required. The
creative condition, which tasked participants with balancing novelty and appropriateness,
engaged a distinct set of brain regions, including the dorsomedial prefrontal cortex and left
inferior frontal gyrus, crucial for higher-order cognitive processes such as planning, decision-
making, and semantic control. This involvement of the left IFG underscores its role in
managing and manipulating semantic information during creative tasks.
In summary, the semantic control network (SCN) is integral to creative thinking,
helping to navigate and reshape semantic memory in the generation of novel ideas. Positioned
between the default and executive networks, the SCN plays a crucial role in balancing
spontaneous and controlled processes, guiding creative output toward relevance and originality.
But, despite its importance, there remains a need for more research to fully understand how

49
semantic control contributes to creativity beyond the general role of associative and executive
processes.

2.7. Thesis Roadmap

In this section, I will present an overview of the work contained within this thesis. The
empirical chapters are presented as self-contained journal articles, prepared for submission or
currently under review. The relevant literature for each study will be presented in the respective
empirical chapter. In the preceding sections, I have argued that semantic control processes play
a key role in creative cognition and have reviewed neuroimaging evidence to support this claim.
Previous literature has established that semantic knowledge is an essential source from which
to draw ideas during the creative process. Moreover, numerous studies have established that
domain-general control mechanisms such as executive functions, fluid intelligence and broad
retrieval ability support creative thinking. However, there are a number of key questions that
have been largely unaddressed in the existing literature.
My thesis addresses these key questions in the following ways. In Chapter 3, I
investigated how semantic memory and semantic control mechanisms, specifically controlled
retrieval and semantic selection, influence creative performance in divergent and convergent
thinking tasks. This study aimed to determine whether these specific semantic control
mechanisms contribute to creative thinking, after accounting for domain-general executive
functions, fluid intelligence and processing speed. Additionally, I explored how age-related
differences in these cognitive abilities effect creative thinking, specifically investigating
whether younger and older adults engage different cognitive mechanisms in the pursuit of their
creative goals, and how these differences impact their performance on divergent and
convergent thinking tasks.
Building on this, I further explored the cognitive mechanisms underlying creative
thought in Chapter 4, focusing on how individual differences in idea generation and evaluation
contribute to performance on divergent and convergent thinking tasks. To this end, I broke
down the classic RAT paradigm into distinct generation and evaluation tasks, processes
intrinsically linked to controlled retrieval and selection mechanisms. This design was adopted
to explore how individuals generate and evaluate ideas when faced with multiple cue
constraints, and whether these abilities relate to creative thinking. Additionally, I conducted a
detailed analysis of the retrieval dynamics during idea generation, examining how components
of semantic search – such as clustering and switching – relate to performance on divergent and

50
convergent thinking tasks. By focusing on these search components, this work sheds light on
the strategies underlying goal-directed retrieval.
In Chapter 5, I shifted focus from traditional divergent and convergent thinking tasks
to the study of spontaneous creativity in speech. Here, I explored creativity as it naturally
occurs in real-world communication, moving beyond the controlled experimental settings of
previous chapters to more ecologically valid assessments of creative cognition. A central focus
here was to examine how speakers balance the need for speech coherence (i.e., structured,
logical communication) with the generation of novel and interesting content. First, I examined
the trade-off between these competing demands at the behavioural level. Next, I investigated
the roles of the DMN, MDN and SCN in supporting these cognitive processes during
spontaneous speech.
Each of these questions guides the empirical work presented in this thesis, providing a
deeper understanding of the cognitive mechanisms that support creative thinking across
different tasks and contexts. In Chapter 6, I present a summary of the findings and discuss
their theoretical implications. I then situate the findings within the broader theoretical
frameworks of Controlled Semantic Cognition and dual-process models of creativity, which
were presented in Chapters 1 and 2 respectively.

51
 Chapter 3

Investigating age related differences in semantic control

mechanisms involved in creative cognition

Abstract

Creative thinking is a complex higher-order ability that draws on multiple cognitive

systems. However, the contribution of specific semantic control processes to creativity
remains unclear. The current study had two goals: First, we investigated how individual
differences in semantic knowledge and control contribute to divergent and convergent styles
of creative thinking, beyond the involvement of domain-general executive functions. Second,
we explored whether there were age-related differences in semantic and executive abilities,
and if these differences influenced the ability to think creatively. Specifically, we examined
the role of the two components of semantic control: controlled retrieval and semantic
selection. In our study, 63 younger adults and 64 older adults completed semantic, executive,
and creative thinking measures. Younger adults demonstrated better executive functioning,
while older adults exhibited superior semantic knowledge, controlled retrieval, and
convergent thinking abilities. Crucially, there were no age differences across several
divergent thinking metrics: automated originality scoring, human ratings and uniqueness.
Regression analyses indicated that semantic knowledge and updating executive ability
influenced convergent thinking abilities across both age groups. In contrast, semantic control
abilities were predictive of divergent thinking skills, but only in the younger group. Our
results emphasize the key role of the semantic system in creative thought, and, critically,
indicate that divergent and convergent thinking may rely on different aspects of semantic
cognition. Moreover, the recruitment of these abilities varies across the lifespan, in line with
increased knowledge reserves and declines in executive control seen in older adults.

Public Significance Statement. The present work demonstrates that older adults can think
just as creatively as their younger counterparts, but these two groups may be relying on
different sets of cognitive abilities. While aging is generally related to cognitive decline, our
results paint a more optimistic picture – knowledge and experience accrued through the
lifespan may be key to the preservation of creative abilities across the life span. These

52
insights highlight the importance of leveraging age-related cognitive strengths to enhance
everyday creative problem-solving.

Acknowledgements. I would like to thank Esperanza Badaya for her invaluable help with
coding the Alternate Uses Task.

Keywords. aging, creative thinking, semantic memory, executive function.

53
3.1. Introduction
Creative thinking is a complex ability that draws from a diverse toolbox of cognitive
mechanisms. Two mechanisms, in particular, are frequently associated with creativity:
divergent thinking, the ability to generate multiple possible responses to an open-ended
constraint; and convergent thinking, the process of homing in on a single solution to a goal-
directed problem (Guilford, 1950; 1967). The literature has often treated divergent and
convergent thinking as opposing forces, with the capability for divergence taken as being more
representative of creative potential (Eysenck, 2003). However, it is likely that both these modes
of thinking are engaged during the creative process, as most creativity tasks require participants
to both produce multiple responses and evaluate them according to task criteria Gabora, 2018).
The interwoven nature of divergent and convergent thinking is echoed in numerous other
theoretical perspectives that suggest that the creative process involves distinct stages of idea
generation and evaluation (Barr et al., 2015; Benedek & Jauk, 2018; Finke et al., 1992;
Kleinmintz et al., 2019; Sowden et al., 2015; Volle, 2018).
As we review below, generation and evaluation rely on different underlying cognitive
mechanisms: idea generation requires the bottom-up activation of concepts from semantic
memory, while top-down executive processing is important for the evaluation and selection of
candidate ideas. The role of semantic representations in the generation of creative thought has
been well established (e.g., Cho et al., 2010; Benedek et al., 2017; Kenett & Austerweil, 2016),
and various studies have also demonstrated a strong link between facets of executive control
and creative abilities (e.g., Beaty et al., 2014; Beaty & Silvia, 2012; Benedek & Neubauer,
2013; Lee & Therriault, 2013; Nusbaum & Silvia, 2011). However, the contribution of specific
semantic control processes, distinct from domain-general executive functions, has not yet been
investigated. Furthermore, there are age-related changes in these cognitive systems: semantic
knowledge and some aspects of semantic control are well maintained through the life span, but
general executive functioning declines (Hoffman, 2018; Kavé & Halamish, 2015; Kavé & Yafé,
2014; Park et al., 2002; Verhaeghen, 2003; Wu, Lohani, et al., 2022). In parallel, recent
evidence suggests that older and younger people may be equally capable of creative thought,
particularly when moderating factors like working memory capacity are considered (Fusi et al.,
2021). This implies that the maintenance of semantic abilities might be important to
understanding the preservation of creativity in later life. Thus, the present study investigated
the contributions of a range of semantic and executive abilities to creative thinking in younger
and older adults.

54
 Semantic memory, which encompasses knowledge about words, objects, abstract
concepts and their properties, is acquired from a collected lifetime of experiences and forms
the basis of many forms of creative thought (Abraham, 2014; Abraham & Bubic, 2015). New
ideas do not simply occur - they are formulated through the manipulation and recombination
of our existing semantic knowledge. Indeed, crystallized intelligence, which indexes semantic
knowledge, is related to both convergent (Ellis et al., 2021; Lee & Therriault, 2013) and
divergent thinking (Cho et al., 2010; Frith et al., 2021). But, while knowledge forms the crucial
foundation for creative thought, the organisation and structure of semantic memory also plays
a vital role.
Mednick’s Associative Theory (1962) highlighted individual differences in semantic
memory structure as being at the root of variability in creative thinking. Mednick defined
creativity as the construction of novel associative links between concepts, with more remote
combinations leading to more creative outputs. Accordingly, the greater the ability to access
and combine disparate concepts, the more creative the individual. Under this framework,
creative individuals are assumed to have ‘flatter’ associative hierarchies, wherein each concept
is linked to a wider range of both common and remote concepts with similar associative
strengths (e.g., ‘shoe’ and ‘home for a family of tiny possums’). In contrast, less creative
individuals possess ‘steeper’ hierarchies characterized by a concentration of a fewer and more
stereotypical associations between concepts (e.g., ‘shoe’ and ‘foot’). Modern computational
network-based approaches that represent semantic memory as a network of associations have
been instrumental in testing Mednick’s theory: numerous studies have examined how the
structure of the semantic system constrains creative thinking at the group (Kenett et al., 2014;
Kenett & Austerweil, 2016) and individual level (Benedek et al., 2017; Luchini et al., 2023).
These studies have demonstrated that more creative individuals have densely connected
semantic networks with shorter distances between concepts and a more flexible structure,
allowing for a faster and more efficient spread of activation (Kenett, 2018; Kenett et al., 2014,
2018; Kenett & Austerweil, 2016; Kenett & Faust, 2019).
Nevertheless, it is not only the structure of the semantic system that constrains the
nature of the associative process. Benedek & Neubauer (2013) administered a divergent
thinking task assessing the degree of uncommonness of word association over time in high and
low creative individuals. They found no differences in the organisation of associative
hierarchies between the groups, but the more creative group was quicker at generating
associations, indicating they may be able to reach more remote combinations at a faster rate.
These patterns of retrieval implicate control processes that go beyond pure association: in

55
addition to the structure of the system, efficiency of access and activation over the semantic
network is also imperative. Further, it is important to note that creativity tasks pose an added
constraint over novelty: they require participants to limit their responses to task-appropriate
ideas. To be creative, participants must inhibit dominant associations, engage in a goal-directed
search for candidate responses, and evaluate them in line with task criteria. Thus, in addition
to a larger, and more densely connected semantic network, creative individuals may be more
skilled in regulating how they access and use this knowledge. These goal-directed retrieval and
selection processes are often referred to as ‘semantic control’ abilities.
Semantic control mechanisms refer to those processes that regulate the retrieval,
manipulation and use of semantic knowledge in a context and task-appropriate way (Badre et
al., 2005; Badre & Wagner, 2007; Hoffman, McClelland, et al., 2018; Lambon Ralph et al.,
2017). Two specific kinds of semantic control process have been proposed: controlled retrieval
and semantic selection (Badre et al., 2005; Badre & Wagner, 2007). Controlled retrieval
processes refer to the goal-directed search through semantic memory when an automatic spread
of activation does not arrive at the desired outcome; and semantic selection processes resolve
competition between multiple competing representations based on current task requirements.
These elements of semantic control may each differentially contribute to creative thought.
During the idea generation phase, controlled retrieval processes may support the formation of
novel conceptual combinations by facilitating access to less common semantic associations.
Meanwhile, semantic selection abilities may be important for inhibiting inappropriate
associations when evaluating candidate responses. The neuroimaging literature supports the
role of the semantic control system in creative thought: brain regions associated with semantic
control have been associated with divergent thinking (Cogdell‐Brooke et al., 2020) and the
generation of unusual links between words (Krieger-Redwood et al., 2023). However, the
relationships between divergent and convergent thinking and measures of the two different
types of semantic control have not been tested at a behavioural level.
Though the role of semantic control in creativity has yet to be fully understood, there
has been much work implicating domain-general executive functions in the regulation and
processing of potential ideas (e.g., Benedek, Könen, et al., 2012; Cheng et al., 2016; Nusbaum
& Silvia, 2011; Palmiero et al., 2022; Zabelina et al., 2012). These domain-general executive
functions are seemingly related to, but distinct from, semantic control processes. At the neural
level, semantic processing has been found to engage domain-general executive control
networks as well as more specialized semantic control regions, particularly those responsible
for retrieval (Bourguignon & Gracco, 2019). Similarly, behavioural evidence indicates a close

56
relationship between semantic selection and general executive abilities, but suggests the
controlled retrieval process is unique to the semantic system (Hoffman, 2018). Thus, semantic
control processes may make a unique contribution to creative thought, beyond domain-general
executive control.
There has been considerable work examining how creative thought is supported by the
three ‘core’ domain-general executive functions proposed by Miyake and colleagues (2000):
inhibition (suppression of dominant but irrelevant information), updating (online monitoring
and manipulating of information in working memory), and shifting (flexibly switching between
relevant task sets). However, a review of 57 studies investigating the relationships between
divergent thinking and the core executive functions found an inconsistent pattern of results
(Palmiero et al., 2022). Most consistently, updating abilities were found to be positively related
to divergent thinking (especially the quantity of ideas) (Batey & Furnham, 2006; De Dreu et
al., 2012; Nusbaum & Silvia, 2011). However, the relationship between divergent thinking and
inhibition was more variable and task dependent, with studies highlighting positive (Benedek,
Franz, et al., 2012; Zabelina et al., 2012) and negative (Cheng et al., 2016; Radel et al., 2015)
associations between the two constructs. This implies that divergent thinking may require
flexible use of inhibitory control depending on task demands. In some tasks, it may be critical
to avoid irrelevant information from contributing to inappropriate ideas, but in others, it may
be more important to allow less salient information to be used to make more distant conceptual
combinations (Vartanian, 2009). Finally, though flexibility has been proposed as being vital to
accessing different semantic categories in the search for original ideas, very few studies have
examined the link between shifting and divergent thinking, and so far, their results have been
inconclusive (Benedek, Jauk, et al., 2014; Zabelina et al., 2019).
Beyond specific executive functions, creativity has also been linked to more general
indices of cognitive ability (e.g., lower order facets of general intelligence such as broad
retrieval ability (Beaty & Silvia, 2012; Forthmann et al., 2019; Miroshnik et al., 2023) as well
as crystallized (Cho et al., 2010) and fluid intelligence (Forthmann, Jendryczko, et al., 2019;
Nusbaum & Silvia, 2011)). Recent meta-analytic evidence indicated that divergent thinking is
correlated with both fluid and crystallized intelligence, further corroborating the idea that both
knowledge and the ability to regulate and evaluate ideas are necessary for creative cognition
(Gerwig et al., 2021).
So far, we have argued that a diverse set of cognitive systems contribute to creative
thought: stored representations of semantic knowledge; specific control processes that act on
these semantic representations; and domain-general executive and cognitive functions. Each of

57
these systems are associated with age-related changes, which may impact creativity in later
life. Older people possess larger stores of knowledge and a rich vocabulary, accrued from their
education and life experience (Kavé & Halamish, 2015; Kavé Gitit & Yafé Ronit, 2014; Park
et al., 2002; Salthouse, 2004; Verhaeghen, 2003; Wu et al., 2022) . Despite this, older adults
have more segregated, less efficient, and less flexible semantic networks (Cosgrove et al.,
2021). These network differences could arise from differences in semantic search abilities
across age groups. Indeed, specific aspects of semantic control are differentially affected by
age; older adults demonstrate preserved controlled retrieval abilities (at least for verbal
knowledge), but are less accurate when the semantic selection demands are high (Hoffman,
2018; Hoffman & MacPherson, 2022; Wu, Lohani, et al., 2022; Wu & Hoffman, 2022).
Additional age-related declines are seen across core executive abilities where older adults
perform worse on tests of executive functioning, exhibiting large magnitudes of differences for
all sub-domains, except for updating (Maldonado et al., 2020). Finally, there are age-related
differences in facets of intelligence, with increased crystallized knowledge and decreased fluid
reasoning abilities (Ryan et al., 2000).
There is little understanding of how age-related changes in these cognitive systems
affect creativity in later life, as most of the research on creativity has focused on young adult
populations and the small body of work looking at creative thinking in aging has found
inconsistent results. Some studies have indicated that the development of creative thinking
across the lifespan assumes an inverted U-shape where divergent thinking abilities peak before
middle age (40-55 years) and then decline (the 'peak-decline' hypothesis; Massimiliano, 2015;
Palmiero et al., 2017; Simonton, 1999). However, more recent evidence suggests that creative
production is preserved in older adults (Madore et al., 2016), especially if mediating variables
like timed vs untimed formats (Foos & Boone, 2008) or working memory (Leon et al., 2014)
are considered. Indeed, a systematic review of 16 papers from 1970-2018 suggests that there
is a complex relationship between aging and divergent thinking (Fusi et al., 2021). This review
highlights the differential effects of aging in verbal and visual domains, with verbal divergent
thinking abilities being preserved in older adults.
In the present study, we collected a wide range of semantic, executive, and creative
thinking measures from younger and older adults. We focused on two key questions. First, we
examined how individual differences in semantic knowledge and control predict creative
abilities, and whether these semantic abilities contribute to creativity beyond domain-general
executive functioning. Second, we investigated variation in semantic and executive abilities
between younger and older adults, and whether these age-related differences can account for

58
performance on creative tasks. To our knowledge, this study is the first to examine how the
specific semantic control components - controlled retrieval and semantic selection -
differentially relate to convergent and divergent thinking, and whether age-related changes in
these abilities affects creative output.

3.2. Methods

3.2.1 Transparency and Openness

In the following section, we have described all information needed to reproduce the
analyses: we have outlined the sample size estimation for the study, as well as all data
exclusions, measures and scoring methods. The analyses were carried out in R ( v4.3.2, R Core
Team, 2023), using the following packages: tidyverse (v 2.0; Wickham et al., 2019), lme4
(v1.1.35.1; Bates et al., 2015), ggeffects (v 1.5.1; Lüdecke, 2018), patchwork (v 1.1.3;
Pedersen, 2023), and wesanderson (v 0.3.7; Ram & Wickham, 2023). The study design and
analyses for this study were not preregistered. Data and code for the study can be accessed at
[Link] and study materials
can be provided upon request. The study received ethical approval from the Psychology
Research Ethics Committee (Ref No: 164-1920/6). In accordance with the ethical requirements,
participants were informed of their rights and informed consent was acquired.

3.2.2. Participants
One hundred and twenty-seven older (n = 64) and younger adults (n = 63) participated
in the study. The older group consisted of adults between the ages of 60-90 years, recruited
through the Volunteer Panel at the Psychology Department, and compensated via a prize draw
for gift vouchers. The younger group included students aged 18-35 years from undergraduate
psychology courses, who were compensated with course credit. All participants were native
English speakers, defined as having learnt English before the age of 5. Participants were asked
to report if they had ever experienced neurological trauma or illness, or a psychological
disorder. Four participants in the Older Adult group were excluded due to a history of
neurological illness, and three participants in the Younger Adult group were excluded due to
colour blindness (n=2) and a diagnosis of Attention-Deficit Hyperactivity Disorder (n=1).
The final sample consisted of 120 people, with 60 people each in the older (41F; MAge =
68.2, SDAge= 5.20) and younger (49F; MAge = 18.7, SDAge = 1.49) groups. A sensitivity analysis
indicated that the minimum correlation between two variables that we could reliably detect
(with 80% power) was r = 0.25 for analyses conducted in each group separately (N=60) and r

59
= 0.18 for analyses of both groups combined (N=120). This is based on a power analysis
conducted in G*Power 3 (Faul et al., 2007), determining that a sample of 60 people in each
group would be sufficient. Data was collected between September 2021 – January 2022 at a
university in the UK.

3.2.3. Procedure
The experiment comprised three parts. In Part 1, participants were tested online, via a
1-1 video call with the experimenter on Zoom. This segment took approximately one hour and
consisted of 6 tasks that were presented to all participants in the following order: (1) Digit
Symbol Matching Task; (2) Series Completion Task; (3) N-back Task; (4) Colour-Word
Interference Task; (5) Alternate Uses Task; and (6) Number Letter Task.
Following the completion of Part 1, participants were asked to complete Parts 2 and 3
online on their own. Both parts were hosted on the Testable website and lasted approximately
20-25 minutes. Part 2 consisted of the semantic tasks: (1) Spot the Word; (2) Synonyms; and
(3) Semantic Control; while Part 3 included: (1) Remotes Associates Task; (2) Kaufman
Domains of Creativity (K-Docs); and (3) Inventory of Creative Activities and Achievements
(ICAA).
The tests used in the study were fully licensed by the University and were initially going
to be administered via face-to-face in-person sessions with the participant. However, due to the
COVID-19 pandemic and associated restrictions, it was necessary to adapt the testing
procedure to an online format, particularly as the study involved a vulnerable older adult
population. To implement this adaptation, the tasks were administered online, in face-to-face
video calls with participants. Materials were not distributed to participants directly. Instead, the
materials were presented to participants via screensharing during the supervised, live video
call. This mimicked standard in-person testing protocols, allowing for control over the test
content and preventing any further access to the content. These modifications were a practical
necessity that we believe fall within the scope of permissible use of the materials.

3.2.4. Measures of Intelligence

We included measures indexing facets of intelligence as outlined in the Cattell-Horn-
Carroll (CHC; Cattell, 1971; Schneider & McGrew, 2018) model, specifically speed of
processing (i.e., the speed of cognitive operations; ‘Gs’) and fluid intelligence (i.e., reasoning
abilities; ‘Gf’). Both speed of processing (Forthmann, Jendryczko, et al., 2019) and fluid
intelligence (Gerwig et al., 2021; Nusbaum et al., 2014; Nusbaum & Silvia, 2011) have been
found to be significant predictors of divergent thinking in the literature, warranting their

60
inclusion as covariates in the current analysis. Importantly, speed of processing is cognitive
ability that is preferentially affected by aging (e.g. Salthouse, 1996), making it particularly
relevant to the present study.

Speed of Reasoning: Digit Symbol Matching Task (DSMT). To measure speed of processing,
the digit symbol matching task from the Wechsler Adult Intelligence Scale-III (Wechsler, 1997)
was adapted to a computerized format. Participants were presented with a key consisting of the
numbers from 1-9, each matched with a corresponding symbol. This key remained on screen
for the duration of the experiment, minimizing any memory demands. Participants were then
presented with one single symbol at a time, in the centre of their screen, and had to match the
symbol to its corresponding number by pressing the correct number key on their keyboard.
There were 90 possible stimulus items, and the final score on the task was calculated as the
total number of correct responses (i.e., correctly matched symbol/number combinations)
generated in a 90-second period. The DSMT has good construct validity, correlating strongly
with other tasks that involve perceptual or processing speed (Salthouse, 2000; Sliwinski &
Buschke, 1999). Additionally, the DSMT has high test-retest reliability (r =.82 -.87) across
different age groups, supporting its stability over time (Wechsler, 1997). Importantly, previous
work using the DSMT has reported substantial correlations with age (ranging from r=-.46 to -
.77; e.g., Hoyer et al., 2004; Salthouse, 1992), supporting its use as a tool to assess age-related
changes in cognitive processing speed.
Fluid Intelligence: Series Completion Task. To measure fluid non-verbal reasoning, we used a
test similar to Raven’s Progressive Matrices (Raven & Raven, 2003): the Series Completion
Task (Scale 3, Form B) from the Cattell Culture Fair tests (Cattell, 1943). Participants were
presented with a series of line drawings that changed according to a certain rule and had to
choose the final image in the sequence. The final fluid reasoning score reflected the total
number of correct responses made within 3 minutes (from a maximum of 13). The Series
Completion Task is a well-validated measure of fluid intelligence, demonstrating strong
internal consistency across all items (α = .91), and high stability over time with test-retest
reliability values ranging from .82 to .85 (Krug, 1973). Regarding construct validity, factor
analytic studies confirm that the Series Completion Task loads strongly onto the general
intelligence (g) factor, aligning it with reasoning and problem-solving abilities (MacArthur,
1962; Cattell, 1968, as cited in Krug, 1973). The task also shows moderate to strong
correlations (r = .51 to .64) with Raven’s Progressive Matrices, supporting its convergent
validity as a measure of fluid intelligence (Wrightstone, 1958; MacArthur, 1963, as cited in

61
Krug, 1973). Given these psychometric properties, the task is highly suitable for assessing
nonverbal problem-solving skills independent of crystallized knowledge.

3.2.5. Measures of Executive Functioning

Three tests were used to measure the core EFs, updating, inhibition and shifting:

Updating: N-back Task. Updating was indexed by the N-back task, based on the version used
by Vaughan et al., (2008). The 2-back version was selected, as a meta-analysis by Bopp &
Verhaeghen, (2018) found that 2-or-more back tests are more sensitive to age effects than the
1-back. Participants were shown a sequence of letters (in white) in the centre of a black
screen. They were told to pay attention to and remember the first two letters of the sequence,
presented at a rate of 2000ms. From the third letter, participants had to respond by indicating
whether the letter they saw currently was the same letter as the one presented two places back
in the sequence. Participants responded by button press, and presentation was self-paced (the
next trial appeared only after the participant responded to the previous trial). There were five
blocks of 20 trials (100 trials in total), presented in the same order to all participants, with a
50:50 ratio of matching to non-matching trials. The average proportion of correct responses
across the 5 blocks were used as an index of performance. The visuoverbal N-back
demonstrates strong test-retest reliability for RT measures of 3-back items (r =.85) and
adequate reliability for accuracy measures (r=.69, Soveri et al., 2018). Unlike static storage
tasks, the N-back primarily measures dynamic working memory updating, as it correlates
weakly with change detection tasks but strongly predicts transsaccadic memory performance
(Frost et al., 2021). Its ability to index online monitoring and maintenance processes makes it
a good measure of working memory updating.

Inhibition: Colour-Word Interference Task. To measure inhibition, we used the Inhibition and
Inhibition/Switching subtests of the Colour-Word Interference test (CWIT) from the Delis-
Kaplan Executive Function System (D-KEFS; Delis et al., 2001), as they show greater age-
related declines than the colour naming (CWIT-1) and word reading (CWIT-2) trials, making
them more sensitive to cognitive ageing effects (Adólfsdóttir et al., 2017). For the Inhibition
subtest, participants were shown a series of colour words written in different coloured ink.
Participants were told to inhibit their dominant response to read out the colour word and to
name the ink colour instead. They were presented with 50 words and the total score was the
total time taken to read all stimulus words. The Inhibition/Switching subtest was similar but
had an additional rule: if there was a box around the word, the participants had to read out the

62
actual word instead of the ink colour. The time taken for participants to respond to 50 stimulus
items was recorded. The final inhibition score was derived by averaging performance across
the Inhibition and Inhibition/Switching subtests, where the higher the score, the poorer the
performance. The CWIT exhibits good psychometric properties: the Inhibition and
Inhibition/Switching subtests demonstrated moderate to high internal consistency using split-
half reliability (r = .72–.90) and moderate to high test-retest reliability (r=.72 and r=.68
respectively; Delis et al., 2001; Homack et al., 2005).

Shifting: Number-Letter Task. The Number-Letter task (Miyake et al., 2000; adapted from
Rogers & Monsell, 1995) was used to index shifting. A square with four quadrants was
presented in the middle of the screen, and a number-letter pair (e.g., ‘G2’) was presented in
one of these four quadrants. The task consisted of 3 blocks: (1) Block 1: The number-letter
combination appeared only in the top quadrants and participants had to indicate whether the
number was odd (3, 5, 7, 9) or even (2, 4, 6, 8); Block 2: The number-letter combination only
appeared in the bottom quadrants and participants had to indicate whether the letter was a
vowel (A, E, I, U) or a consonant (G, K, M, R); (3) Block 3: The number-letter combination
first appeared in the top left quadrant and then rotated through each quadrant in a clockwise
direction. If the number-letter combination was in the top quadrants, participants had to
respond to the number (i.e., say whether the number was odd or even) but if it was in the
bottom quadrants, participants had to respond to the letter (i.e., say whether the letter was a
vowel or a consonant). Participants were given 10 practice trials of each block. Blocks 1 and
2 consisted of 32 trials each and did not require task switching; Block 3 consisted of 64 trials,
and the trials in the top left and bottom right quadrants required switching between number
and letter tasks (32 trials). The shifting cost was calculated as the difference between the
average RTs for switch trials in Block 3 and the average RT for no-switch trials in Blocks 1
and 2. The Number-Letter task has been widely used as a measure of task-switching ability
and demonstrates strong psychometric properties. Test-retest reliability is moderate to high
for switch trials (r = .73) and for switching costs (r = .68; Soveri et al., 2018). Internal
consistency is also high, with Cronbach’s α = .91 for switch trials and α = .92 for non-switch
trials (Friedman et al., 2016). In terms of validity, the task correlates well with other measures
of shifting, including the Colour-Shape Task (r = .41) and the Category-Switch Task (r = .48),
supporting its construct validity as an index of cognitive flexibility and mental set shifting
(Friedman et al., 2016).

63
3.2.6. Measures of Semantic Ability
Semantic Knowledge: Spot the Word Task. To evaluate the breadth of participants’ store of
semantic information, they completed the Spot the Word (STW) test from the Speed and
Capacity of Language Processing battery (Baddeley et al., 1992). In the STW test (60 trials),
participants were shown a pair of response options (e.g., toaster-flumpter) consisting of a
word, and a phonologically and orthographically plausible nonword, and had to pick the real
word. The STW has good internal reliability (α = .83), and correlates strongly with
performance on the Mill Hill vocabulary scale (r = .86) as well as the National Adult Reading
Test (NART; r = .60), indicating that it is an effective index of verbal knowledge (Baddeley et
al., 1993). Additionally, it is important to consider potential limitations of the SPT related to
reliance on non-semantic processing strategies. Specifically, the task could be completed
using whole-word orthographic recognition instead of semantic memory itself, i.e.,
participants may be able to identify familiar letter patterns or word shapes rather than
accessing the meaning of the words. However, this is unlikely to be the predominant strategy
across all items. The design of the STW test includes words of varying frequencies, from
common to highly infrequent (Baddeley et al., 1993). For many items, mere orthographic
recognition is insufficient for accurately distinguishing between real words and non-words,
and as the difficulty level increases, more specific word knowledge is needed. Moreover, the
STW test correlates with other tests of verbal knowledge (Baddeley et al., 1993; Yuspeh &
Vanderploeg, 2000), suggesting that semantic knowledge is a key component of performance.
Therefore, while the test may not be a pure measure of semantic processing, it is a reliable
indicator of crystallized intelligence, appropriate for our purposes in the current study.

Semantic Knowledge: Synonyms Task. To test the breadth of semantic knowledge, we used a
version of the Synonyms task adapted from the Mill Hill vocabulary test (Hoffman, 2018;
Raven et al., 1989). On the Synonyms task (44 trials), the participants were presented with a
probe word (e.g., run) and four response options (turn, dash, sigh, expire) and had to match the
probe with its synonym (dash). Following previous studies (Hoffman, 2018; Hoffman &
MacPherson, 2022), the percentage correct responses on the Spot the Word Task and Synonyms
Task were combined to form a composite score indexing semantic knowledge.

Semantic Control Tasks. Participants were tested on the two forms of semantic control, using
tasks that have been previously employed in the literature (e.g., Badre et al., 2005; Hoffman,
2018; Whitney et al., 2011; Wu & Hoffman, 2022). Global semantic association trials were
used to probe participants’ ability to engage in controlled retrieval of less salient semantic

64
information. Participants had to choose the semantic associate of a cue word (e.g., pebble) from
a set of four alternatives (e.g., letter, cat, stone, person). This association was manipulated to
be either strong (e.g., town-city) or weak (e.g., iron-ring), with the weak associations requiring
more controlled retrieval of appropriate semantic information to identify the relevant semantic
link. There were 24 strong and 24 weak trials over the four blocks of global trials. The
proportion of correct responses on the weak trials was taken as a measure of controlled retrieval
ability.
To measure semantic selection, we used feature-based association trials that examined
the extent to which participants could resolve competition between multiple semantic
alternatives (Thompson-Schill et al., 1997). Participants were presented with a probe (e.g.,
pepper) and four response options (e.g., ice, tar, salt, cabbage). They had to match the probe
word with a target based on a particular item property (here, colour). On congruent trials, the
probe and target were matched on the relevant feature and had a strong semantic association,
placing low selection demands on participants as the automatic activation of semantic
associations would prompt the correct response (e.g., strawberry-raspberry). On the
incongruent trials, the probe and target did not have any pre-existing association (e.g., pepper-
tar), requiring participants to focus specifically on feature-relevant attributes shared by the cue
and each of the possible responses. These incongruent trials included foils which were strongly
related to the target word (pepper-salt), placing further demands on semantic selection
processes by necessitating the inhibition of dominant but irrelevant knowledge. Participants
were informed which feature they should be focusing on at the start of each block (size or
colour). There were 24 congruent and 24 incongruent trials over the 4 blocks of feature-based
trials and the proportion of correct responses on the incongruent trials was taken as a measure
of semantic selection ability.

3.2.7. Measures of Creativity: Self Report

While our primary interest in the study was the relationship between cognitive abilities to
performance-based measures of creative thinking, we included two self-report measures of
creativity to assess the degree to which the effects transferred to creative behaviours in
everyday life.

Creative Potential: Kaufman Domains of Creativity (K-Docs). The K-Docs (Kaufman, 2012)
is a self-report measure that probes beliefs about one’s own creative abilities. The K-Docs
presents participants with a list of 50 creative behaviours, covering different domains such as
Self/Everyday, Scholarly, Performance, Mechanical/Scientific, and Artistic creativity.

65
Participants are asked to rate their creative ability for each act in comparison to their peers,
on a 5-point Likert scale from 1 ‘much less creative’ to 5 ‘much more creative’. The possible
scores for this measure range between 50 – 250, where the higher the score, the higher the
creative abilities. The K-Docs exhibits strong internal consistency, with a pooled Cronbach’s
α of .904 (Sen & Yörük, 2023), and test-retest reliability coefficients ranging from .76 to .86
across subscales (Kaufman, 2012).

Creative Abilities: Inventory of Creative Activities and Achievements (ICAA). The ICAA
(Diedrich et al., 2018) is a comprehensive evaluation of a person’s real-life engagement in
various domains of creativity, including literature, music, arts and crafts, creative cooking,
sports, visual arts, performing arts and science and engineering. This self-report measure
includes an assessment of both creative activities (little-C) and creative achievements (Pro-
C). The Creative Activities (CAct) scale indexes the frequency with which the person has
undertaken a given activity over the last 10 years. Each of the eight domains contained six
items and responses were made on a 5-point Likert scale (0=never; 1 = 1-2 times; 2 = 3-5
times; 3 = 6-10 times; 4 = more than 10 times). The maximum total possible score for
creative activities is 240, with a higher score indicating a higher frequency of creative
activities. The Creative Achievements (CAch) scale presents participants with various levels
of attainment in each domain, each relating to an increasing value from 0 to 10 (e.g., from 0 =
“I have never been engaged in this domain” to 10 = “I have already sold some work in this
domain). The first 5 levels cover personal achievements (from “I have tried this domain
once” to “I have already taken classes to improve my skills”) while the remaining levels
cover public achievements (“I have already published my original work in this domain” to “I
have already sold work in this domain”). Participants are required to indicate all possible
levels of attainment that apply to them (multiple answers are permitted). A domain-general
creative achievements score is calculated by summing the pertinent levels of attainment
across all domains. For each of the eight domains, the scores may range from 0 to 55, with a
maximum possible score of 440, indicating the highest level of creative achievements. The
ICAA demonstrates good internal consistency (α = .92; Diedrich et al., 2018). The CAct
subscale demonstrates internal consistency ranging from α = .86 to .89 across studies, with
test-retest reliability over a 4-week period at r = .81. The CAch subscale also exhibits strong
reliability (α = .88), though test-retest reliability is slightly lower (r = .67 over 4 weeks).

66
3.2.8. Measures of Creativity: Convergent and Divergent Thinking Tasks
Convergent Thinking: Remote Associates Task. To examine convergent thinking abilities, the
Remote Associates Test (RAT; Mednick, 1962) was administered. Each trial consists of three
cue words (e.g., bald – screech – emblem), each of which are linked to a target word (here:
eagle) that participants must try to identify. The cue words were presented in the centre of the
screen with a blank text box underneath, and participants were given 30 seconds to type their
response in the box. Thirty trials were presented, 15 of which were taken from Mednick’s
(1962) original items and 15 of which were developed for a previous study by our lab (See
Open Science Framework:
[Link] Responses were
corrected for spelling errors, and nonsensical words, incomplete responses, and timed out
responses were classified as inadequate responses and scored as incorrect. Responses that
were very close synonyms to target words were accepted and marked as correct responses
(e.g., sea-ocean). The final score for this task reflected the total number of correct responses.
The original RAT demonstrates high internal reliability, with Spearman-Brown reliability
coefficients reported as r = .92 and r = .91 in two separate samples (Mednick, 1962).
Regarding validity, RAT scores correlate moderately to highly with measures of intelligence
and problem-solving, but more weakly with divergent thinking tasks, suggesting it primarily
assesses convergent thinking processes (Lee et al., 2014; Mednick, 1962).

Divergent Thinking: Alternate Uses Task. The Alternate Uses Task (Guilford, 1967) was used
to assess divergent thinking. Participants were given the name of an everyday object and were
asked to generate creative uses for the object in 2 minutes. The task was audio-recorded, and
participants were asked to verbalize their responses out loud. As our main focus was the
creative quality of the responses, the participants were instructed to ‘be as creative/original as
possible’, in line with recent recommendations regarding measurement and scoring of
divergent thinking (Beaty et al., 2022; Reiter-Palmon et al., 2019). Accordingly, we explicitly
instructed them to focus on two dimensions of creativity, namely originality and
appropriateness. Finally, they were told to prioritize quality over quantity via the following
instruction: “Come up with as many ideas as you can, but it’s more important to be creative
than to come up with a lot of ideas” (Nusbaum et al., 2014). Participants were given 2 cue
words (hanger, bucket), and were given 2 minutes per item.

67
3.2.9. Scoring the Alternate Uses Task
All responses were transcribed by the first author. For each item, responses were inspected
for adequacy according to the guidelines in Guilford’s AUT test manual (Forthmann et al.,
2016; Wilson et al., 1960). Accordingly, responses related to selling, borrowing or gifting
were marked as inadequate. Further, common uses for each object, repetitions, and vague,
incomplete or nonsensical ideas were also excluded and marked as inadequate. There are
multiple methods of obtaining creativity metrics from the AUT, both empirical and
subjective. Meta-analytic findings highlight concerns about discriminant validity amongst
different methods of scoring the AUT, namely fluency, flexibility and orginality (Acar et al.,
2023). Similarly, Benedek (2024) emphasized that while the AUT remains a central tool for
assessing creative potential, its validity is influenced by design choices in measurement and
scoring. A primary issue is the fluency confound, where higher response quantity (fluency)
artificially inflates originality scores, making it difficult to determine whether individuals are
truly generating more creative ideas or simply producing more responses overall. However,
despite these concerns, appropriate scoring methods can effectively mitigate fluency-related
biases and distinguish between distinct components of divergent thinking (Acar et al., 2023).
Given these considerations, we incorporated multiple creativity metrics—both empirical and
subjective—to ensure a comprehensive evaluation that captures different facets of divergent
thinking and minimizes potential biases associated with any single approach. Here, we use
four of the most common measures: response fluency, automated scoring of response
originality, subjective human ratings of creative quality, and response uncommonness in the
sample. We computed all four metrics to evaluate the extent to which they provided similar
indices of divergent thinking, with higher scores signifying better performance.

Fluency. To index the productivity aspect of divergent thinking, fluency scores were computed
for each item by summing up the total number of adequate responses, and then averaging this
across the two items.

Originality. To compute a measure of creative quality, responses were scored for originality
using open-source software developed by Organisciak et al. (2022). This tool uses supervised
machine-learning to score AUT responses. It was developed by fine-tuning a neural network-
based large-language model constructed with GPT-3 architecture (Brown, 2020), by training it
on past responses to the AUT. Across a large collection of datasets and prompts, Organisciak
et al. (2022) found their model to be extremely strong at predicting human ratings of creativity
(r=.81). Using the model, each response was scored on a scale of 1-5, where 1 was very

68
unoriginal, 5 was very original. An overall originality score was calculated by averaging
responses for the two items.

Uniqueness. To index creative quality, we used an empirical measure of uncommonness based

on the statistical frequencies of the response in the study sample (Forthmann et al., 2020). First,
responses were categorized according to the degree of feature overlap, with functionally
analogous uses grouped together (Reiter-Palmon et al., 2019). For example, possible uses of a
bucket as a ‘shopping bag’ and ‘book bag’ would be grouped together as functioning to ‘carry
items’. Next, a response occurrence table was built based on the number of equivalent
responses that were identified and the frequency of each distinct response was estimated. An
independent rater reviewed the categorization of a subset of the responses and expressed
agreement with the coding scheme. Weights were assigned according to fixed threshold values
of 1% and 5% (Runco, 2008). Responses that were provided by 1% of the sample were
considered as highly unique and assigned a score of 2; those that were reported by less than 5%
of the sample were considered somewhat unique and assigned a score of 1. All other responses
were assigned a value of zero. Due to the potential confounding effects of fluency, Uniqueness
scores for a participant’s responses were averaged rather than summed (Forthmann,
Szardenings, et al., 2020). A total Uniqueness score for each participant was derived by
averaging their uniqueness score across the two items.

Subjective Creativity Ratings. To measure the creative quality of responses, subjective ratings
of creativity were obtained from 80 native English-speaking individuals with no prior
experience. They were given some background information about the AUT and told that they
would be asked to rate responses generated by participants from a previous study. Each
response would be rated on its creative quality, on a scale of 1 (not at all creative) to 5 (highly
creative). The raters were provided with a set of instructions adapted from previous work
(Silvia et al., 2008) that highlighted key dimensions of creative output: uncommonness;
remoteness; and cleverness. Further, raters were told that strength in one of these dimensions
could balance weakness in another dimension (e.g., a relatively common use that is expressed
in a unique way could receive a higher score). The full instructions are provided in
Supplementary Materials (see Appendix A). Raters undertook some practice trials before they
started with the main task. For each AUT item (i.e., hanger and bucket), equivalent responses
and repetitions were grouped together (e.g., ‘as a plant pot’ and ‘planter’), while responses that
were unique in the way they were worded or expressed (e.g., if they were more descriptive)
remained individual items. For each item, two subsets of responses per age group were

69
constructed, with each subset containing approximately 175-200 responses. Each subset was
rated by 10 individuals who took approximately 25 minutes to complete the ratings. In order
to standardise how responses were scored across subsets, each rater’s scores were normalised
using the midpoint of the Likert scale (i.e., 3). The total creativity rating for a given response
across raters was computed as the average of these normalised scores.

Elaboration. Elaboration refers to the length of each response given by a participant. Although
this is not a measure of creativity itself, we included this metric as automated methods of
originality scoring may be biased by the number of words in a response. For example, responses
with more words have lower LSA-derived semantic distance values (Forthmann, Oyebade, et
al., 2019; Forthmann, Szardenings, et al., 2020). Moreover, human raters are also influenced
by the specific way in which the idea is expressed, and tend to rate longer responses as more
creative (Beaty & Johnson, 2021). To account for this, elaboration of the response was scored
using an unweighted word count approach using whitespace tokenization (i.e., counting the
number of words in the response), which has been shown to be a reliable measure (Dumas,
Organisciak, Maio, et al., 2021).

3.2.10. Statistical Analyses

To explore the relationships between the various cognitive abilities, Pearson’s
correlations were computed. To remove age group effects, the data were z-scored within each
group prior to calculating the correlations. A principal components analysis of the creativity
metrics was also performed on the z-scored data, and factors with eigenvalues greater than one
were extracted and varimax rotated.
To examine whether semantic and executive abilities contribute to age differences in
convergent thinking, and whether these predictors varied by age group, multilevel logistic
regression models were run on RAT accuracy scores. Sum to zero coding was adopted for the
Group factor, with Older Adults as the first level. To account for individual and trial-level
variation, these models included random by-participant and by-trial intercepts. The model
building approach was as follows: our base model included only group and semantic abilities
as predictors, and in a second step, our full model also included measures of executive
functioning (updating, inhibition, shifting) and broad cognitive ability (speed of reasoning,
fluid intelligence). We also tested whether the effects of cognitive predictors differed between
age groups. To do this, we iteratively compared the full model with models that included an
interaction between group and each of the individual predictors using Likelihood Ratio Tests.
If the addition of a group-level interaction term improved model fit, it was included in the final

70
model. Continuous predictors were standardized prior to their inclusion in the model. Models
were fit using maximum likelihood estimation and the bobyqa optimiser from
the lme4 package (Bates et al., 2015). P values were calculated using the lmerTest package
(Kuznetsova et al., 2017), in which the degrees of freedom are approximated using
Satterthwaite’s method.
Multiple regression models were run to investigate the contribution of semantic and
executive abilities to performance on the AUT metrics (originality, uniqueness, subjective
ratings, and fluency), and whether this varied by age group. Given that there were only two
AUT trials, multiple regression was chosen over mixed effects models, as there were not
enough observations to estimate random effects. Elaboration has been established as an
important confound in AUT scoring (Beaty & Johnson, 2021; Forthmann, Oyebade, et al.,
2019), and so was included as a predictor in all models. Initially, two models were run for each
of the AUT outcomes: in the base model, the AUT metric was regressed on age group, semantic
abilities, and elaboration; the full model additionally included all executive and broad cognitive
functions as predictors. Following this, the incremental F-tests were used to iteratively compare
the full model to models with an interaction between age group and each of the individual
predictors. If the group-level interaction term improved model fit, it was retained in the final
model. Finally, if there was a significant group-level interaction, these effects were followed
up with separate models in each age group, examining the effects of the semantic and executive
abilities on each AUT metric.

3.3. Results

3.3.1. Age differences on measures of divergent thinking, semantic knowledge, and control
mechanisms
The means and standard deviations of the variables of interest are shown in Table 1. Younger
adults outperformed their older counterparts on tests of executive functioning (Inhibition,
Switching and Updating), speed of reasoning (Digit Symbol) and fluid intelligence (Series
Completion). On the other hand, older adults demonstrated better semantic knowledge
(lexical decision and synonym judgement), and performed more highly on the controlled
retrieval task, but not semantic selection. Regarding the creativity measures, older adults
were better at convergent thinking, with higher scores on the Remote Associates Task (RAT)
than the younger group. Importantly, the two groups displayed similar performance on the
divergent thinking measures: older and younger adults had similar scores for subjective
ratings (AUT ratings), uniqueness (AUT uniqueness) and originality (AUT originality).

71
However, older adults were more fluent (AUT fluency) than the younger group. On self-
report measures of creativity, older adults reported engaging in more creative activities, while
the younger group reported more creative achievements. In summary, while the older group
exhibited poorer domain general executive abilities, they had an advantage on some semantic
tasks and convergent thinking, and their divergent thinking abilities were on a par with the
younger group.

Older adults completed significantly more years of full-time education (15.75 years,
SD = 2.48) than younger adults (13.28 years, SD = .88), t (73.71)= 7.09, p <.001. Although
the groups did differ in the years of education undertaken, this was not considered as a
covariate in the final analysis. Instead, we focused on tests of semantic knowledge tests that
were more directly relevant to creative performance, as they specifically assessed the depth of
semantic knowledge required for successful task completion. For example, proficiency on the
RAT requires possession of the relevant knowledge, making word knowledge a more
appropriate predictor. Additionally, education was not considered as a covariate in the final
analysis, as this difference was expected due to our sampling technique. By design, the
younger group, being first-year undergraduate students, had fewer years of education.

72
 Table 1: Descriptive Statistics by Age Group. Mean (M) and Standard Deviation (SD) values
are reported for each group.
Ability Task Units Younger adults Older adults

Fluid Intelligence Series Completion Total correct 7.35 (1.69) 6.25 (1.80)***
(max = 13)

Speed of Reasoning Digit Symbol Total correct in 90s 49.52 (3.68) 37.62 (5.29)***
(max = 90)

Inhibition Colour-Word Interference Time to completion (s) 44.17 (5.90) 56.87 (11.46)***

Updating N-back task Proportion 0.94 (0.04) 0.88 (.09)***

Correct

Shifting Number-Letter Task RT switch minus RT no 812.60 (482.07) 1294.15 (786.47)***

switch (ms)

Spot the Word Lexical Decision % correct 78.94 (7.67) 93.94 (5.06)***

Synonyms Synonym Judgement % correct 37.5 (11.74) 71.55 (12.59)***

Knowledge Lexical decision & synonym % correct 58.22 (8.37) 82.75 (8.14)***
judgement

Controlled Retrieval Weak Associates Proportion 0.84 (.07) 0.91 (.06)***

Correct

Semantic Selection Incongruent Feature Proportion 0.82 (.13) 0.85 (.12)

Selection Correct
Convergent Thinking Remote Associates Test Total correct (max = 3.58 (.13) 6.87 (2.38)***
30)
Fluency Alternate Uses Test No. of responses 7.43 (2.99) 9.63 (3.75)***

Uniqueness Alternate Uses Test Uncommonness of 0.47 (.23) 0.46 (.24)

responses
Creativity Ratings Alternate Uses Test Subjective ratings 3.48 (.11) 3.52 (.12)

Originality Alternate Uses Test AI-based scoring 2.75 (.19) 2.79 (0.19)

Elaboration Alternate Uses Test No. of words 9.22 (3.16) 10.35 (3.72)

Creative Self-Beliefs K-DOCS Self-report (50-250) 147.22 (21.75) 145.76 (15.05)

Creative Activities ICAA Self-report 65.70 (24.17) 78.37 (17.50)***

(max = 240)
Creative Achievements ICAA Self-report 71.22 (47.83) 27.97 (20.49)***
(max = 440)
Note: K-DOCS = Kaufman Domains of Creativity; ICAA = Inventory of Creative Activities and Achievements. Standard deviations are shown in
parentheses. Asterisks indicate the significance of t-tests comparing younger and older adults (Older adults column). * = p <.05; ** = p <.01; *** = p
<.001.

73
3.3.2. Relationships between creativity, semantic and executive abilities
Pearson’s correlations were used to investigate the relationships between creativity,
semantic and executive abilities. Figure 1 shows the relationships between measures in the full
dataset. Scores were normalized within each group prior to computing correlations, so these
effects are independent of age group differences.

Figure 1: Pearson correlations between creativity, semantic and executive abilities in the
combined dataset. Note: Inhibition and Shifting used time-based measures so higher values
indicate poorer performance (unlike the other measures), RAT = Remote Associates Test;
AUT = Alternate Uses Test. * = p <.05; ** = p <.01; *** = p <.001.

With respect to the semantic abilities, knowledge and controlled retrieval correlated with
one another, but not with semantic selection. This aligns with previous studies indicating that
semantic selection is a distinct process from other aspects of semantic cognition (see Hoffman,
2018; Wu & Hoffman, 2022). Both semantic control measures were associated with updating,
but not with the other executive and general cognitive measures. Of the core executive control
measures, shifting was related to both inhibition and updating, but updating and inhibition were

74
not related. Speed of reasoning correlated with all three core executive functions, while fluid
intelligence was strongly related to updating.
As expected, the different AUT metrics were strongly inter-correlated, and the self-report
measures were correlated with one another. However, the RAT was somewhat distinct, as it
correlated weakly with AUT originality but not with the other creativity measures. To explore
the latent relationships between the different measures of creativity, exploratory principal
components analyses were performed on the z-scored data. The Minimum Average Partial
(MAP) method and Parallel Analysis suggested that a 2-factor solution provided the best fit for
the data, with the 2 components together accounting for approximately 46% of the variance in
scores in each of the age groups (see Table 2). The first component appeared to index creative
quality, as AUT metrics of uniqueness, human creativity ratings, originality, and elaboration all
loaded on this factor. The second component appeared to index creative engagement, as it was
loaded on by the self-report measures of creative behaviours and achievements, as well as the
number of responses given in the AUT (Fluency). On the other hand, the RAT did not load
strongly on either of these components, suggesting it may tap into a distinct dimension of
creative thinking.

Table 2: Principal Components Analysis of creativity measures.

Component 1 Component 2
RAT .20 -.26
AUT Fluency .05 .62
AUT Uniqueness .57 .26
AUT Ratings .78 -.09
AUT Originality .87 .10
AUT Elaboration .60 -.04
Creative Self-Beliefs .05 .68
Creative Activities -.05 .71
Creative Achievements .30 .66
Note: RAT = Remote Associates Task; AUT = Alternate Uses Task.

Factor loadings > 0.4 are shown in bold.

With regard to creativity measures and other cognitive functions, the RAT was correlated
with updating, semantic knowledge and controlled retrieval, implying that there may be some
specific contributions of both semantic abilities and executive control to convergent thinking.

75
In contrast, there was a more complex pattern of correlations with the AUT measures,
suggesting that different aspects of executive and semantic control are associated with specific
elements of the AUT. Finally, self-reported creativity was not strongly associated with
executive and semantic functions.
Separate correlational analyses for each age group are shown in the supplementary
materials (Appendix A, Figure S1). While many of the associations were replicated across age
groups, there also appeared to be some striking differences between age groups. For example,
semantic selection ability was correlated with AUT measures in the younger group but not the
older group. Overall, these patterns of correlations indicate that multiple cognitive abilities may
be associated with performance on the AUT and the RAT. To examine this further, we
conducted regression models to understand which tests independently predicted these scores,
when controlling for shared variance, and whether these relationships varied between age
groups.

3.3.3. Predictors of performance on RAT

Table 3 shows the effects in mixed models predicting RAT scores. In the first model,
including semantic knowledge and control abilities as predictors, there was a strong effect of
knowledge but no independent effects of controlled retrieval or semantic selection. Executive
functions and general cognitive abilities were added in the second model, which significantly
improved the model fit, as demonstrated by a likelihood ratio test (χ2(5) = 11.81, p<.05). The
effect of knowledge remained highly significant in the second model; after controlling for
executive abilities, the odds of correctly answering RAT items increased by 1.55 with every 1
SD increase in semantic knowledge. Updating was also a significant predictor: higher N-back
performance predicted better RAT scores, with a 1.19 increase in RAT performance for every
1 SD increase in updating abilities. Effects of age group were not significant in either model,
suggesting that older people’s advantage on the RAT is largely attributable to their more
developed semantic knowledge. Finally, we examined whether any of the effects of cognitive
scores varied between age groups, but no test x group interaction terms improved the model
fit.

76
Table 3: Effects of age, semantic and executive functions on Remote Associates Task (RAT)
performance.
Effect Model 1 Model 2
OR CI P OR CI p
Group 1.17 0.96 – 1.43 0.130 1.18 0.92 – 1.52 0.184
Knowledge 1.49 1.21 – 1.83 <.001 1.55 1.27 – 1.90 <.001
Controlled Retrieval 1.09 0.95 – 1.25 0.203 1.04 0.91 – 1.19 0.584
Semantic Selection 1.02 0.91 – 1.14 0.766 1.01 0.90 – 1.12 0.904
Updating 1.19 1.04 – 1.37 0.013
Inhibition 1.09 0.95 – 1.25 0.197
Shifting 1.10 0.97 – 1.24 0.140
Fluid Intelligence 1.09 0.97 – 1.23 0.148
Speed of Reasoning 1.01 0.82 – 1.25 0.892

3.3.4. Predictors of performance on AUT

Table 4 demonstrates the effects in regression models predicting AUT Originality

scores. The first model included only the semantic predictors, none of which significantly
predicted originality. AUT elaboration was also included as a covariate in all AUT models, and
as expected, more elaborate responses received higher originality scores. The final model
included the executive functions, general cognitive abilities and interactions of abilities with
age group (wherever these improved model fit). The final model was a better fit for the data,
as demonstrated by an incremental F test, F(7)=2.40, p=0.024. There were no main effects of
test scores but there was a significant interaction between group and semantic selection, and
elaboration remained a significant covariate. The interaction effect is shown in Figure 2. We
investigated this interaction by estimating separate models for each age group (See Table 1 in
Appendix C). There were no semantic control effects in the older group; however, in the
younger group, after controlling for executive abilities, participants with better semantic
selection ability produced more original AUT responses (β = 0.07, 95% CI [0.02-0.11], p<.01).
Group level results are presented in Appendix A (Table S1).

77
Table 4: Analyses of effects of age, semantic and executive functions on the Alternate Uses
Task (AUT) Originality.
Effect Model 1 Model 2
B (se) CI p B (se) CI p
Group -0.02 -0.08 – 0.04 0.470 0.03 -0.04 – 0.10 0.440
Semantic Selection 0.03 -0.00 – 0.06 0.057 0.02 -0.01 – 0.05 0.150
Knowledge 0.02 -0.04 – 0.07 0.565 0.01 -0.05 – 0.07 0.705
Controlled Retrieval 0.02 -0.02 – 0.06 0.311 0.01 -0.03 – 0.05 0.568
AUT Elaboration 0.07 0.04 – 0.11 <.001 0.08 0.05 – 0.11 <.001
Updating 0.02 -0.02 – 0.06 0.304
Inhibition 0.01 -0.03 – 0.05 0.684
Shifting 0.02 -0.02 – 0.05 0.298
Fluid Intelligence 0.03 -0.01 – 0.06 0.140
Reasoning Speed 0.05 -0.01 – 0.10 0.126
Group × Semantic -0.04 -0.07 – -0.01 0.006
Selection
Group × Controlled -0.03 -0.06 – 0.01 0.157
Retrieval

Figure 2: Two-way interaction plot of the predicted effects of semantic selection on AUT
Originality scores. Note: OA = Older Adults, YA = Younger Adults. The shaded areas
represent the 95% confidence intervals.

78
 Table 5 shows effects in models predicting AUT Ratings scores. In the first model,
none of the semantic predictors had an effect on human ratings of creativity, but AUT
elaboration was a significant covariate. The inclusion of executive functions, general cognitive
abilities and interaction terms in the final model improved model fit, as demonstrated by an
incremental F test, F(9)=4.80, p <.001. In the final model, after controlling for executive
abilities, there were significant interactions between group and semantic selection, controlled
retrieval and speed of reasoning. Further, elaboration remained a significant covariate and there
was a main effect of inhibition, where people with poorer inhibition abilities producing
responses that were rated as being more creative. These interactions are shown in Figure 3 and
were investigated further in separate group level models. For younger adults, after controlling
for executive abilities, subjective human ratings of creative quality were predicted by semantic
selection (β = 0.03, 95% CI [0.01-0.06], p = .005) and controlled retrieval abilities (β = 0.04,
95% CI [0.01-0.06], p = .004) (See Appendix A, Table S2). However, this was not the case for
the older group, wherein speed of reasoning (β = 0.07, 95% CI [0.03-0.11], p = .001) and
inhibition (β = 0.04, 95% CI [0.00-0.07], p = .032) significantly predicted ratings of creativity.

Table 5: Analyses of effects of age, semantic and executive functions on Alternate Uses Task
(AUT) Ratings.
Effect Model 1 Model 2
B (se) CI p B (se) CI P
Group 0.00 -0.03 – 0.04 0.891 0.04 -0.00 – 0.08 0.075
Semantic Selection 0.01 -0.01 – 0.03 0.295 0.01 -0.01 – 0.03 0.454
Knowledge 0.00 -0.03 – 0.04 0.864 -0.00 -0.03 – 0.03 0.980
Controlled Retrieval 0.01 -0.01 – 0.03 0.356 0.01 -0.01 – 0.03 0.503
AUT Elaboration 0.04 0.02 – 0.06 <.001 0.05 0.03 – 0.07 <.001
Reasoning Speed 0.07 0.03 – 0.10 <.001
Updating -0.00 -0.03 – 0.02 0.877
Inhibition 0.03 0.00 – 0.05 0.030
Shifting -0.00 -0.02 – 0.02 0.948
Fluid Intelligence -0.01 -0.03 – 0.01 0.272
Group × Semantic -0.02 -0.04 – -0.01 0.008
Selection
Group × Controlled -0.03 -0.05 – -0.01 0.011
Retrieval
Group × Reasoning 0.04 0.01 – 0.07 0.023
Speed
Group × AUT -0.02 -0.04 – 0.00 0.067
Elaboration

79
Figure 3: Effects of cognitive abilities on AUT Ratings. Note: OA = Older Adults, YA =
Younger Adults. The shaded areas represent the 95% confidence intervals.

Table 6 displays the effects in models predicting AUT Uniqueness scores. The first
model indicated that semantic selection was a significant predictor of Uniqueness. However,
this effect disappeared when accounting for executive and general cognitive abilities. In the
final model, there were no main effects or interactions with age group. Group level results are
presented in Appendix A (Table S3).

80
Table 6: Analyses of effects of age, semantic and executive functions on Alternate Uses Task
(AUT) Uniqueness.
Effect Model 1 Model 2
B (se) CI p B (se) CI p
Group -0.04 -0.12 – 0.04 0.300 0.03 -0.07 – 0.12 0.616
Semantic Selection 0.05 0.01 – 0.10 0.016 0.04 -0.00 – 0.08 0.080
Knowledge 0.04 -0.04 – 0.11 0.364 0.02 -0.06 – 0.10 0.554
Controlled Retrieval -0.01 -0.06 – 0.04 0.757 -0.02 -0.07 – 0.04 0.517
AUT Elaboration 0.03 -0.01 – 0.08 0.118 0.03 -0.01 – 0.08 0.128
Updating 0.04 -0.01 – 0.10 0.150
Inhibition -0.05 -0.10 – 0.01 0.088
Shifting -0.01 -0.06 – 0.04 0.653
Fluid Intelligence -0.00 -0.05 – 0.04 0.945
Reasoning Speed 0.00 -0.08 – 0.08 0.957
Group × Semantic -0.03 -0.08 – 0.01 0.113
Selection
Group × Controlled 0.04 -0.01 – 0.09 0.141
Retrieval

Table 7 shows effects in models predicting AUT Fluency scores. While the first model
indicated that semantic selection was a significant predictor of Fluency, this effect was no
longer present when accounting for executive and general cognitive functions in the final
model. However, in the final model, there was a main effect of age group.

Table 7: Analyses of effects of age, semantic and executive functions on Alternate Uses Task
(AUT) Fluency.
Effect Model 1 Model 2
B (se) CI p B (se) CI p
Group 0.87 -0.24 – 1.98 0.122 1.56 0.11 – 3.01 0.035
Semantic Selection 0.67 0.06 – 1.28 0.032 0.46 -0.16 – 1.07 0.145
Knowledge -0.02 -1.14 – 1.10 0.969 -0.27 -1.40 – 0.86 0.632
Controlled Retrieval 0.42 -0.30 – 1.14 0.247 0.41 -0.34 – 1.15 0.285
AUT Elaboration -0.25 -0.87 – 0.37 0.420 -0.16 -0.79 – 0.46 0.605
Fluid Intelligence 0.39 -0.26 – 1.05 0.235
Reasoning Speed -0.08 -1.29 – 1.14 0.898
Updating 0.30 -0.50 – 1.10 0.459
Inhibition -0.32 -1.12 – 0.49 0.438
Shifting -0.30 -1.01 – 0.40 0.396
Group × Fluid 0.64 -0.05 – 1.34 0.069
Intelligence

81
3.4. Discussion

The present study sought to understand age-related changes in divergent and convergent
thinking, and the specific contributions of semantic and executive abilities to creative thought
in younger and older adults. As expected, younger adults in our study exhibited better domain-
general executive functioning, speed of reasoning and fluid intelligence. However, the older
group outperformed their younger counterparts on tests of semantic knowledge, controlled
semantic retrieval and convergent thinking. Crucially, the two groups displayed equivalent
divergent thinking abilities across several metrics, including AI-based originality scoring,
human creativity ratings and uncommonness of responses. An exploratory principal
components analysis suggested that divergent and convergent thinking were separable
constructs, and only the ability to generate original responses on the Alternate Uses Test was
weakly correlated with performance on the Remote Associates Test. This dissociation was
echoed in the regression analyses where mixed effects models indicated that semantic
knowledge and updating skills contributed to convergent thinking across age groups. In
contrast, multiple regression models suggested that these factors were less relevant for
divergent thinking. Instead, semantic control mechanisms predicted divergent thinking abilities
in the younger adults, while more general cognitive abilities like speed of reasoning and
inhibition predicted performance in older adults. Taken together, these results suggest that
divergent and convergent thinking may draw on distinct sets of cognitive abilities, and that
there are age-related differences in the recruitment of these abilities.
Our older adults exhibited deficits in executive functioning, echoing the findings of
previous studies (Argiris et al., 2020; Bopp & Verhaeghen, 2018; Idowu & Szameitat, 2023;
MacPherson et al., 2002; Maldonado et al., 2020; Rey-Mermet & Gade, 2018; Salthouse, 2009;
Verhaeghen & Salthouse, 1997; Wasylyshyn et al., 2011). On the other hand, our older group
exhibited greater depth of semantic knowledge, in line with numerous previous studies
chronicling the increase in stores of semantic knowledge with age, accrued through years of
education and lifelong literary, cultural and workplace experience (Hoffman, 2018;
Verhaeghen, 2003). In addition, our older group also demonstrated superior convergent
thinking skills, related to their larger stores of semantic knowledge. In accordance with
previous work (Hoffman, 2018; Hoffman & MacPherson, 2022), we found that older adults
also exhibited a greater ability to engage in the controlled retrieval of less salient semantic
associations. However, this advantage did not extend to our older adults’ performance on the

82
semantic selection task, replicating previously established age-group dissociations between
these two aspects of semantic control (Hoffman, 2018; Hoffman & MacPherson, 2022).
However, while our sample of older adults performed numerically more poorly on the semantic
selection task, the difference between age groups was not statistically significant (unlike
Hoffman, 2018). As our sample was younger than that of Hoffman (2018), it is possible that
age differences in this task are more pronounced when much older participants are included.
Across groups, breadth of semantic knowledge was correlated with controlled retrieval
abilities, indicating that individuals with a larger base of semantic information develop more
effective search mechanisms. Nevertheless, similar to Hoffman (2018) and Wu & Hoffman
(2022), semantic selection was not related to the other semantic abilities, providing further
support for the distinction between these two types of semantic control. Finally, both controlled
retrieval and semantic selection performance were related to updating, demonstrating the
importance of being able to hold and manipulate multiple pieces of information in mind when
engaging in strategic semantic retrieval and selection between competing alternatives.
For convergent thinking, people with larger stores of semantic knowledge and better
updating abilities demonstrated better performance on the RAT, after controlling for semantic
control and other executive abilities. To solve a RAT item, individuals must hold multiple cues
in mind while searching through the intersection of their semantic neighbourhoods for the
related target word, following which they must engage in a sequential evaluation of candidate
responses (Smith et al., 2013). Greater reserves of conceptual knowledge may provide an
individual with a larger search space for the bottom-up spreading of activation processes during
the idea generation phase. This is in line with previous research indicating that crystallized
intelligence, a measure that taps vocabulary and verbal intelligence, uniquely predicts multiply-
constrained problem solving tasks, including the RAT (Ellis et al., 2021). While our study
provides further support for the necessity of semantic knowledge for verbal convergent
thinking tasks, the structure and organization of semantic memory may also be relevant when
making connections between distant concepts. Recent research using network science
demonstrated that individuals who performed well on the RAT had richer and more flexible
semantic networks (Luchini et al., 2023). These were similar to the semantic networks of
individuals with strong divergent thinking skills (Benedek et al., 2017; Kenett & Faust, 2019),
indicating that both modes of thought may rely on the flexible organization of conceptual
information. Finally, semantic knowledge alone is not sufficient to solve RAT problems; we
found that updating, the ability to manipulate the contents of working memory, also predicted
performance. This aligns with previous work that has highlighted the importance of working

83
memory capacity for multiply-constrained problem solving tasks like the RAT (Ellis & Brewer,
2018; Ellis et al., 2021). Top-down executive control allows for the maintenance and
manipulation of the three cues in working memory while searching through our conceptual
knowledge reserves and comparing candidate solutions to the features of the cue words.
In the AUT, our test of divergent thinking, younger and older people were similar in
terms of their performance on measures of creative quality indexed by originality, subjective
ratings, and uniqueness. However, older people generated more responses, and were more
elaborate in their descriptions of the object uses. These results are in accordance with previous
studies suggesting that, despite declines in executive abilities, verbal divergent thinking
abilities are preserved in older adults (Fusi et al., 2021; Madore et al., 2016). The present study
is the first to highlight potential differences in the mechanisms underlying creative performance
in younger and older adults. Despite similar levels of performance, the predictors of AUT
success varied significantly between age groups, with semantic control abilities being more
important for performance in the younger group. Controlled retrieval abilities predicted
subjective ratings of creative quality and uniqueness; and semantic selection abilities seemed
to influence originality scores, subjective ratings, and uniqueness, even after accounting for
general executive functions. The AUT requires the generation of multiple uncommon, but
appropriate, uses for a common object. Enhanced controlled retrieval abilities facilitate this
process by allowing fewer common associations between concepts to be activated, leading to
the generation of novel uses. At the same time, improved semantic selection abilities afford
individuals the ability to inhibit dominant uses and filter out less promising associations,
resulting in better and more original responses. Thus, together these processes enable more
efficient generation of original yet task-appropriate ideas.
Although better semantic control abilities were associated with improved performance
on AUT for younger people, this effect was not present in the older group. Taken together, our
results suggest that, while older and younger adults displayed equivalent divergent thinking
abilities, these groups may differentially draw on various cognitive processes to achieve the
same outcome. Younger adults, with less experience of the world, may rely on their semantic
control systems to leverage their limited knowledge flexibly, creating new connections
between distant concepts as they seek to generate uses. Older adults, on the other hand, possess
large stores of conceptual knowledge and life experience, which may allow them to retrieve
suitable uses more automatically from their existing knowledge base. This could include
retrieval of episodic memories of object-use events that they have previously experienced or
witnessed. In this age group, the limiting factor could be the more general regulation of

84
responses and filtering out of less creative ideas, which could account for the effects of speed
of reasoning and inhibition seen in the older group.
Our AUT results are in line with the default-executive coupling hypothesis of aging
(DECHA; Spreng & Turner, 2019) which proposes that, with age, there is a shift toward
semanticized cognition (i.e., an increased reliance on prior knowledge, including semantic,
conceptual and schematic information), coupled with declines in cognitive control. The authors
suggest that this reliance on prior knowledge leads to a shift in cognitive mode (Hills, Todd,
Lazer, et al., 2015) from exploratory (novelty-seeking, preference for external environment) to
exploitative (novelty-aversive, reliance on prior knowledge). At the neural level, this is
reflected in adaptive functional changes in network connectivity; there is greater coupling
between the prefrontal cortex and default network in older people, perhaps indicative of greater
reliance on stored knowledge (Spreng & Turner, 2019). This is also seen in brain-based
investigations of creativity, wherein older adults demonstrate reduced functional connectivity
within the default and executive control networks but greater connectivity between the two
(Adnan et al., 2019). Together, this suggests that older adults may be offsetting losses in
executive control by using their wealth and variety of experiences to maintain their creative
abilities. This tendency was also apparent in the RAT analysis, where the older group’s more
advanced semantic knowledge appeared to account for their greater task success, despite their
poorer updating ability.
It is important to note that, unlike the RAT, simply having a larger store of knowledge
did not confer any benefits to performance on the AUT. Convergent thinking tasks like the
RAT have multiple cue constraints and require individuals to arrive at a specific target
response. This cannot happen unless individuals already possess the specific information
required (i.e., the relevant relationships between each of the cue words and the target), thus,
conferring an advantage on individuals who have a wider knowledge base. On the other hand,
divergent thinking tasks are more flexible and open-ended, and can be approached in many
ways. Here, what an individual knows may be less critical than how efficiently they can identify
and use their knowledge in a novel way. Finally, it could also be that our knowledge measure
did not capture the most relevant forms of knowledge for the AUT. We operationalized
knowledge as verbal vocabulary knowledge. While this may be important for the RAT, it may
not be as relevant for the AUT, which could depend instead on non-verbal object knowledge
and more general life experiences.
We used a number of metrics to score responses in the AUT: subjective human ratings
of creative quality; AI-based scoring of originality; and a measure of statistical uncommonness

85
in the sample. While all the AUT metrics were correlated, the strong association between the
AI-based originality scoring and human ratings underscores the potential of automated scoring
approaches as a viable alternative to the more time consuming, human-based scoring (Buczak
et al., 2023; Organisciak et al., 2023). Importantly, all the AUT metrics loaded on a single
factor indexing creative quality in our exploratory principal components analysis. So far, the
literature has poised divergent and convergent thinking as separate elements of creativity, and
associations between the two have been inconsistent (Lee et al., 2014; Lee & Therriault, 2013).
Our principal components analysis supports this view; the RAT and the AUT metrics tapped
different underlying constructs. However, we did find that the RAT was weakly related to AI-
based originality scores. Finally, it is worth noting that only originality and human ratings were
predicted by cognitive abilities – this may be because uniqueness is a more crude and less
sensitive measure of divergent thinking, and requires very large samples to accurately estimate
how uncommon a given response is in the population (Forthmann, Paek, et al., 2020; Reiter-
Palmon et al., 2019).
While the present study provides valuable insights into the cognitive processes
underlying divergent and convergent thinking in older and younger adults, there are a few
methodological limitations worth noting. One such limitation concerns the number of items
used in the AUT. Presenting participants with 2 items for the AUT may have limited the range
of creative ideas expressed, as it may lead to stimulus dependency. Intra-individual
performance can vary significantly across different items, with participants showing
preferences for one stimulus over the other (Barbot, 2018). This preference may be influenced
by factors such as the extent of the participant’s prior experience with the object (Runco et al.,
2006) or the salience of the object itself (Forthmann et al., 2016). Moreover, the AUT
commonly uses items like typical household objects which are highly familiar to participants,
which may in turn elicit more conventional response patterns (Beaty et al., 2022). Indeed,
varying the objects used and number of trials administered could help mitigate biases related
to stimulus characteristics and improve on the reliability of creativity measures in the AUT
(Barbot, 2018).
However, despite these concerns, prior work has suggested that even with a limited
number of items, the AUT can still capture meaningful variations in creativity. In our study,
we opted for a simpler design, using two items per participant, to balance participant
engagement and task demands, especially considering the older adults’ potential for cognitive
fatigue. Given that the older adults were asked to engage in multiple cognitive tasks, including
tasks of executive control and semantic retrieval, we felt that using two stimulus items would

86
reduce fatigue while still providing reliable data. By limiting the number of items, we aimed to
ensure that participants could fully engage with each task without fatigue or cognitive overload.
Furthermore, by employing a range of different metrics to extract creativity indices from the
data, we were able to ensure a comprehensive evaluation of participants’ creative abilities was
still achieved.
Another limitation of the present study concerns the characteristics of the sample,
particularly regarding the older adult cohort. Older adults reported being engaged in
significantly more creative activities than the younger group. This may be due, in part, to the
fact that many older adults in the sample were retired, providing them with more time and
opportunity to engage in creative pursuits. In contrast, the younger group, who were full-time
undergraduate students, had more time constraints with balancing education, employment and
other responsibilities. Additionally, retired older adults may have accumulated more financial
resources that enable them to participate in hobbies and classes, opportunities that may be less
accessible to younger adults with limited disposable income. Thus, the observed differences in
creativity activity engagement between the two groups could partially reflect disparities in time
and resources. Moreover, it is important to note the potential for selection bias as our sample
of older adults was recruited through the Psychology Department’s Volunteer Panel. This panel
likely consists of individuals with a predisposition for participating in cognitive studies,
potentially over over-representing those who are more actively engaged in creative or
intellectual pursuits. Thus, this may not reflect broader age-related trends in creativity.
To conclude, the current study demonstrated that semantic knowledge contributes to
convergent thinking in younger and older people, but that these groups may differentially
recruit semantic and executive abilities for divergent thinking. These findings are in line with
neurocognitive theories of aging. Moreover, our study provided further evidence that older
people are as successful at divergent thinking tasks as their younger counterparts. While aging
is related to cognitive decline across several areas, especially those related to cognitive control,
age also brings a wealth of knowledge and life experience that helps older people manoeuvre
through daily life. This knowledge seems to be especially relevant for older adults to engage
in creative thinking to find new ways to adapt when navigating everyday problem-solving.

87
 Chapter 4

Dissecting the RAT: Exploring the Dynamics of Semantic Search

in Creative Thinking

Abstract

Creative ideas often stem from a deliberate process involving the exploration and
exploitation of our existing knowledge. This study explores the cognitive mechanisms behind
creative problem-solving, focusing on the phases of idea generation and evaluation. Generating
novel ideas relies on an effective, goal-directed search through semantic memory, while
evaluation requires the careful assessment of potential ideas for their appropriateness. These
phases are analogous to semantic control processes, which manage the search, manipulation
and selection of knowledge in line with specific goals. The current work investigates how
individual differences in these cognitive processes relate to performance in divergent and
convergent thinking tasks, by deconstructing the standard Remote Associates Test (RAT) to
isolate generation and evaluation abilities. Further, we examined the role of retrieval strategies,
specifically clustering (grouping related concepts) and switching (moving between different
areas of semantic space), in creative thought. The findings indicate that the ability to generate
ideas under multiple constraints predicts fluency in divergent thinking tasks, while strong
evaluation skills may inhibit originality. Furthermore, the analysis of clustering and switching
strategies reveals that effective semantic search involves a balance between deep exploration
of specific conceptual areas and broad exploration across different semantic clusters. These
findings suggest that multiple strategies can lead to creative success, underscoring the
importance of flexible cognitive strategies in navigating the semantic space during creative
problem-solving.

Acknowledgements. I would like to thank Georgia Carter and Melissa Thye with their help in
coding responses to the Alternate Uses Task.

88
4.1. Introduction
Creative ideas rarely appear out of nowhere – Eureka! moments aside, creative thought
requires the effortful exploration and exploitation of our existing knowledge. When engaged
in creative problem-solving, individuals search their knowledge for relevant ideas and evaluate
these candidate ideas in line with task constraints. Rather than being a sequential process,
creative thinking may be underpinned by two iterative phases: bottom-up idea generation that
draws on the contents of our semantic and episodic memory; and top-down idea evaluation
that employs various control processes to assess the quality of the proposed ideas (Kleinmintz
et al., 2019). The distinct phases of idea generation and evaluation are ostensibly analogous to
semantic control processes underlying the search, manipulation and selection of knowledge in
pursuit of task-related goals (Badre et al., 2005; Hoffman, 2018; Lambon Ralph et al., 2017).
They are also linked with two creative thinking modes: divergent thinking (generation of
multiple response options) and convergent thinking (homing in on one ‘correct’ solution;
Guilford, 1950; 1967). However, despite these apparent parallels, researchers have not
examined how individual differences in the ability to generate and evaluate candidate ideas
predict performance in divergent and convergent thinking tasks.
The ways in which individuals search through their semantic space when generating
new ideas may also have a bearing on how creative they are. When searching for concepts,
people typically retrieve multiple semantically-related concepts from an area of semantic space
(clustering) before transitioning to another area (switching). Individual differences in these
retrieval dynamics may contribute to efficient navigation of semantic space, resulting in the
generation of more creative ideas. Nevertheless, the contributions of these search components
to creative cognition have rarely been studied. To these ends, the present study had two main
aims: first, we examined how the ability to generate and evaluate ideas was related to creative
thinking; and second, we conducted a detailed investigation of the idea generation process,
testing how individual differences in clustering and switching strategies relate to creative
performance.
The structure and organization of semantic knowledge is fundamental to idea
generation (Abraham, 2014; Abraham & Bubic, 2015). Associative theories of creativity
emphasise individual differences in semantic memory structure and the spontaneous,
associative processes operating over our knowledge stores (Beaty & Kenett, 2023; Kenett &
Faust, 2019; Mednick, 1962). Under this framework, information is stored as a network of
associations of varying strengths, and creative thinking results in the construction of new ideas
by linking previously unconnected, remote concepts. This conceptualization is akin to

89
spreading activation theories of semantic memory, which posit that semantically similar items
are stored closer together and so, are activated more easily (Collins & Loftus, 1975). Thus,
highly creative individuals are characterised as having a semantic memory structure with
numerous, weakly linked concepts, which allows for the avoidance of dominant, stereotypical
associations in favour of retrieving more remote ideas. Recent computational work has
supported and extended this theory using network-based approaches to demonstrate that more
creative individuals possess densely connected, flexible semantic networks with shorter path
lengths between concepts that support a more efficient spread of activation (Kenett, 2018;
Kenett et al., 2014, 2018; Kenett & Austerweil, 2016; Kenett & Faust, 2019).
While the content and organization of our semantic knowledge provides the foundation
for creative thought (e.g., Abraham, 2014; Abraham & Bubic, 2015; Benedek et al., 2017;
Kenett et al., 2014; Luchini et al., 2023), controlled search processes are also vital for the
efficient access and manipulation of this information. Even when individuals possess similar
associative networks, more creative individuals arrive at remote associations at a faster rate,
highlighting the importance of flexible and efficient access over the semantic network
(Benedek & Neubauer, 2013). In line with this view, recent theoretical work has recognized
the role and importance of top-down executive control in creativity (e.g., Beaty et al., 2014;
Benedek & Fink, 2019; Silvia, 2015), and studies have underscored the importance of various
control processes for creative thought, including domain-general executive functions (e.g.,
Benedek et al., 2012; Cheng et al., 2016; Palmiero et al., 2022; Zabelina et al., 2012), fluid and
crystallized intelligence (e.g., Cho et al., 2010; Forthmann, Jendryczko, et al., 2019; Nusbaum
& Silvia, 2011), and other facets of intelligence such as broad retrieval ability (Beaty & Silvia,
2012; Forthmann, Jendryczko, et al., 2019; Miroshnik et al., 2023). However, while much work
has focused on the contribution of these broader executive abilities to creativity, questions
remain about how more specific semantic control abilities relate to the creative process.
Studies in cognitive neuroscience have illustrated the role of control processes in
regulating access to and use of semantic knowledge. These semantic control mechanisms
regulate the search, manipulation, and use of semantic information in a task-appropriate way
(Badre et al., 2005; Badre & Wagner, 2007; Hoffman et al., 2018; Lambon Ralph et al., 2017;
Reilly et al., 2023). Specifically, two forms of semantic control have been suggested: controlled
retrieval processes that work to inhibit dominant associations and facilitate access to more
remote and novel ideas; and semantic selection processes that resolve competition between
candidate solutions in line with task requirements (Badre et al., 2005; Badre & Wagner, 2007).
These control abilities map onto the dual process framework that considers creativity in terms

90
of idea generation and evaluation. Indeed, neuroimaging studies do suggest that semantic
control regions are activated during various forms of creative thought (Cogdell‐Brooke et al.,
2020; Krieger-Redwood et al., 2023). However, little is known about how performance on
tasks of divergent and convergent thinking is related to individual differences in these abilities
to generate candidate ideas from semantic memory and evaluate them for their appropriateness.
It is also important to consider cognitive processes operating within the generation
phase, as there remain unresolved questions about how retrieval strategies may contribute to
success in different creativity tasks. When generating creative ideas, the automatic spreading
of activation between highly related concepts may fail to avoid stereotypical response patterns,
prompting the engagement of goal-directed retrieval strategies. Two search components –
namely, clustering and switching – have been implicated in semantic fluency tasks, as well as
in the creative process at a behavioural and neurocognitive level (Mastria et al., 2021; Ovando-
Tellez et al., 2022). Clustering (i.e., producing groups of semantically associated words in
sequence) has long been observed in studies of memory retrieval: when engaging in semantic
fluency tasks, individuals tend to exhibit a pattern of identifying, and then further examining,
sub-categories of concepts or ‘semantic fields’, as termed by Gruenewald & Lockhead, (1980).
A distinct switching component was first proposed in the cluster-switching hypothesis (Troyer
et al., 1997), which emphasized the role of transitioning to other sub-categories once a
particular sub-category is exhausted. This emergent clustering and switching behaviour is
thought to stem from a local-to-global search process, present in many two-stage models of
memory retrieval, such as the Search of Associative Memory Model (Raaijmakers & Shiffrin,
1981). Typically, the first stage involves the local retrieval of similar items, which is followed
by a global exploration for a new cluster. This iterative cycling between local and global modes
is also characteristic of foraging models, which liken semantic search to the way that animals
search for food in ‘patchy’ environments, with patch resources depleting during the foraging
process (Hills, Todd, & Jones, 2015). According to the optimal foraging theory (Hills et al.,
2012), when the goal is to maximize foraging gains, there is a trade-off between exploiting a
current patch and leaving the patch in search of a more resource-rich area. This ‘exploration-
exploitation’ trade-off may be applied to how humans search for information in patches of
semantic space by tending to switch between patches when the retrieval rate in the original
patch falls below a certain threshold.
Most work on retrieval dynamics in semantic search has used semantic fluency tasks,
where the participant has only one task constraint: to search for all members of a specific
category. However, creative thinking tasks often involve multiple constraints. For example, in

91
the Remote Associates Test (RAT; Mednick, 1962), a classic test of convergent thinking,
individuals are provided with three cue words and asked to find a target word that relates to all
three cues. Multiply-constrained problems such as these are thought to involve an iterative two-
stage process: search through the semantic space of each cue word and their intersection, and
testing of responses against the cue words for acceptability of the solution (Smith et al., 2013).
But, as most studies using the RAT only capture a single response from the end of the search-
test mechanism, questions remain about what happens behind the scenes of this problem-
solving process.
Some studies have broken the search-test process down to examine the specific search
processes underlying performance on the RAT. Smith and colleagues (2013) gave participants
two minutes to solve a RAT problem, asking them to record all their possible candidate
solutions generated during their search for the right answer. Rather than demonstrating the clear
clustering/switching patterns seen in semantic fluency tasks, participants’ response patterns
were reflective of a sequential dependence (i.e., each response was semantically similar to the
preceding response, even across cluster breaks). The authors suggested that these results were
indicative of a local search algorithm that allowed for a more efficient exploration of the
semantic space. Conversely, Davelaar (2015), using a similar paradigm, found pronounced
clustering/switching behaviour - specifically, people tended to cluster less and switch more.
While all three cues activated their own cue-specific semantic areas, local search through these
areas did not result in finding the solution. Instead, the optimal semantic search occurred in the
intersection of the semantic neighbourhoods of all three cues. Davelaar suggested that this anti-
clustering behaviour (i.e., switching from patch to patch by activating cues sequentially) was
key to finding the target solution on the RAT. In response, Smith and Vul (2015) argued that
clustering does occur in RAT search but that, rather than being a cognitive strategy, it might be
an epiphenomenal consequence of sequential dependence, which they framed as a fundamental
feature of semantic search. If this sequential dependence is not considered, responses may
appear as being clustered when in fact response generation is only constrained by the preceding
response.
To sum, the few studies that have investigated semantic search in multiply constrained
RAT problems do not agree on any one optimal search strategy. Instead, the literature puts forth
two possible predictors of success on the RAT: sequential dependence and anti-clustering
behaviour. However, it is important to note that these prior studies focused on clustering and
switching patterns in the generation of candidate solutions, but did not investigate how

92
individual differences in these search components are related to performance on more
conventional divergent and convergent thinking tasks.
More broadly, the literature on individual differences in clustering and switching, and
their relationship to creative abilities, is sparse. Ovando-Tellez et al. (2022) examined
clustering/switching behaviour on a free association task cued by polysemous words with
multiple meanings. The two search components were differentially related to performance on
the RAT, and the Alternate Uses Task (AUT; Cortes et al., 2019; Reiter-Palmon et al., 2019), a
commonly used measure of divergent thinking. Clustering was strongly associated with the
number of responses produced on the AUT, while switching was related to accuracy on the
RAT. Similarly, Zhang et al. (2023), using clustering/switching metrics derived from a chain-
free association task, demonstrated a link between these search components and both real-life
creativity and performance on divergent thinking tasks. However, both the ambiguous word
task and the chain-free association task index search behaviour related to only one cue
constraint, and search processes may be very different in tasks with multiple cue constraints
(Davelaar, 2015; Smith et al., 2013).
In the present study, we investigated how performance on commonly-used measures of
convergent and divergent thinking (RAT and AUT) are predicted by people’s (a) efficiency in
idea generation and evaluation and (b) clustering/switching behaviours. With this objective in
mind, we deconstructed the Remote Associates Test, a standard measure of convergent
thinking, to isolate generate and evaluation ability. The Generation-RAT was similar to the
version used by (Smith et al., 2013) and involved participants generating as many candidate
solutions to RAT problems as they could, without needing to evaluate their accuracy. In
contrast, in the evaluation-RAT, participants were given four possible solutions to a RAT
problem and simply had to choose which one was correct. We used performance on these two
tests as measures of generation and evaluation, and further, analysed responses in the
Generation-RAT to quantify participant’s propensity to cluster and switch during semantic
search. Further, we examined whether the propensity to cluster and switch during semantic
search contributes to creative thinking on tasks with multiple cue constraints.

4.2. Methods

4.2.1. Participants
The initial sample consisted of 190 participants. Forty-eight participants were recruited
from Testable Minds, an online participant pool hosted on the Testable website, and received
monetary compensation for their participation. One hundred and forty-two undergraduate

93
psychology students from the University of Edinburgh were recruited from the School of
Philosophy, Psychology and Language Sciences research participation pool and compensated
with course credit. Participants were included if they fulfilled the following criteria: learned to
speak English before the age of 5 years; showed a minimum level of engagement by (1)
producing at least an average of two responses on the AUT, and (2) producing an average of
five responses on the Generation-RAT. Based on this criteria, we excluded 40 participants who
reported learning to speak English after the age of 5 years; six participants who generated less
than two responses on average on the AUT; and 27 participants who produced less than five
responses on average on the Generation-RAT. Our final sample consisted of 110 participants
(M Age = 20.08, sd Age = 3.67, range = 18-38; 69 Female, 37 Male, 4 Other).

4.2.2. Procedure
This was a self-administered, online behavioural study hosted on the Testable website.
Participants were asked to compete five tasks in the following order: (1) Standard Remote
Associates Task; (2) Alternate Uses Task; (3) Generation-RAT; (4) Evaluation-RAT; (5)
Creative Behaviour Inventory. Example trials from each task are shown in Figure 1. The entire
experiment lasted approximately one hour.

4.2.3. Tasks
Standard Remote Associates Task. To measure convergent thinking, we used the Remote
Associates Test (Mednick, 1962; Wu et al., 2020). On a classic RAT problem, the participant
is presented with three cue words and must identify a target word that links to all three cues.
Using the cues as constraints, participants must search their semantic knowledge and make
novel connections to home in on the correct answer. Participants were given 15 of Mednick’s
(1962) original items and 30 seconds per trial. Cue words were presented on the centre of the
screen, with a blank text box underneath. Spelling errors were rectified, and nonsensical or
incomplete responses were excluded. Close synonyms to target words were marked as correct
(eg., sea-ocean). The final score was recorded as the total number of correct responses out of
15 items.

Alternate Uses Task. To index divergent thinking, we used the classic Alternate Uses Task
(Guilford et al., 1978). Participants were asked to generate novel uses for an everyday object.
In line with the recommendations (Hocevar, 1979), participants were asked to be ‘as
creative/original’ as they could. Specifically, they were instructed to generate responses that
were both original and appropriate, and asked to focus on quality over quantity of responses.

94
Participants were given two items (chair, belt). For each item, they were presented with a blank
text box and were given two minutes to type out their responses.

In accordance with Guilford’s AUT test manual, the following kinds of responses were
marked as inadequate and excluded from the analysis: responses related to selling, borrowing,
gifting; the object’s common uses; repetitions; and vague, incomplete, and nonsensical ideas.
We employed three common AUT metrics: fluency, originality, and uniqueness. Fluency,
representing an individual’s capacity for generation, was computed as the total number of
adequate responses per trial. Next, an AI-based originality scoring method (Organisciak et al.,
2022) that is strongly correlated with human ratings of creativity (r = .81) assessed creative
quality. The model scored each response on a 1-5 scale, with higher scores reflecting more
original responses. To compute an overall originality score, we averaged over the scores for
each response to a given trial. Finally, a measure of uniqueness based on the statistical
frequency of the response in the sample assessed creative quality. Functionally analogous uses
were categorized together, according to their degree of feature overlap. Then, a response
occurrence table was constructed to estimate the frequency of each distinct category of
response. Fixed thresholds of 1% and 5% were utilized to weight responses: ‘highly unique’
responses generated by less than 1% of the sample were scored 2, ‘somewhat unique’ responses
generated by less than 5% of the sample were scored 1, and all other responses received a score
of 0. To account for potential fluency confounds (Forthmann, Szardenings, et al., 2020), we
divided the uniqueness for a given trial by the number of responses, and then computed an
overall uniqueness score by averaging across the two items. We also included a measure of
elaboration (i.e., the length of each response given by a participant) as this has been shown to
bias the AI-based originality scoring (Forthmann, Oyebade, et al., 2019; Forthmann, Paek, et
al., 2020). Elaboration was scored as the number of words in the response (Dumas et al., 2021).

Generation-RAT. To index controlled retrieval of semantic associates, we used a modified form

of the RAT (Davelaar, 2015; Smith et al., 2013). Participants were provided with three cue
words (e.g., ‘clump’, ‘weight’, ‘public’), and asked to type out every candidate response that
they considered while searching for a solution that related to all three words. Participants were
provided with 2 minutes per trial. Unlike previous studies, we instructed participants to
continue generating candidate responses even if they thought they had found the correct
answer. This was done to measure participants’ ability to generate a greater variety of possible
responses, and to avoid underestimating the generation abilities of participants who found the
solution more rapidly. Further, participants were asked not to use proper nouns. Ten trials were

95
administered to all participants in the same order. The total number of words generated was
taken as the final G-RAT score.

Evaluation-RAT. To measure the ability to evaluate candidate ideas, we devised a novel

adaptation of the classic RAT. In this version, participants were provided with three cue words
(e.g., ‘lower’, ‘kid’, ‘frivolous’) and were presented with three possible candidate solutions
(e.g., ‘airport’, ‘degrade’, ‘minor’). Their task was to select the target word that was related to
all three cues (here, ‘minor’). Thirty items for the E-RAT were designed and piloted in a sample
of 30 Native English speakers, recruited through the researchers’ personal network (MAge =
27.64, SDAge = 2.07). Items that had an accuracy rate of 40% or more were chosen for the final
study. Twenty trials were administered to all participants in the same order. The final E-RAT
score reflected the total number of correct responses out of 20 items.

Creative Behaviours Inventory. The Revised Creative Behaviours Inventory (CBI; Dollinger,
2003), a shortened version of Hocevar's (1979) scale, was used to index everyday creative
activity. Participants were presented with a list of 28 common creative activities or
achievements (e.g., writing a poem). They were asked to indicate how frequently they had
engaged with this behaviour in their life on four-point scale (Never, 1-2 times, 3-5 times, more
than 5 times). The scale has good reliability (Cronbach’s alpha ranging from .88 to .92) and
validity (see Silvia et al., 2012). While our focus was on examining individual differences in
the ability to generate and evaluate and their relationship to performance on creative thinking
tasks, we included the CBI as a self-report measure of creativity to assess the degree to which
these abilities were relevant to creative behaviour in daily life.

96
Figure 1: Schematic Illustration of Tasks. (a) Remote Associates Task; (b) Alternate Uses
Task; (c) Generation RAT; (d) Evaluation RAT; (e) Response sequence generated for an
example G-RAT problem. Note: Cue 1: insult; Cue 2: shear; Cue 3: interrupt.

4.2.4. Measures of Clustering/Switching Behaviour

For responses to G-RAT problems, metrics of semantic similarity were calculated

between each response and each cue word, and each response with the preceding response.
These were used to determine clusters and switches, as illustrated in Figure 2. Spelling errors
were corrected, and responses were excluded if they were incomplete or nonsensical, or if they

97
contained the cue words. Semantic similarity was calculated using embeddings from the
word2vec neural network trained on the Google News corpus (Mikolov et al., 2013). In this
model, words are represented as high-dimensional vectors with words that are used in similar
ways (and thus are similar in meaning) displaying more similar vectors. During the training
phase, the model learns to predict words based on their surrounding context, allowing for the
capture of syntactic and semantic relationships. The degree of relatedness between a pair of
words was computed using the cosine similarities between their vectors, ranging from -1
(completely dissimilar) to 1 (identical).
Based on the semantic similarity, each response was assigned to the ‘primary’ cue word
it was most related to, following the procedure outlined by Smith and colleagues (2013).
Beginning with the second response in the chain, each response was labelled ‘cluster’ if it was
assigned to the same primary cue as the previous response; otherwise, it was labelled as a
‘switch’ (see Figure 2). These labels were used to determine each participant’s
clustering/switching characteristics. In the literature, various metrics have been used to
characterise clustering/switching behaviour (e.g., Ovando-Tellez et al., 2022; Troyer et al.,
1997). We computed a series of these metrics and then used principal components analysis
(PCA) to identify the latent factors that underpinned these behaviours. Each of these metrics
was calculated at the trial level and then averaged across all G-RAT trials.

Mean number of switches. When a response was assigned to a different cue than the previous
response, this was defined as a switch. For each trial, we counted the total number of switches.
We then divided this by the total number of responses minus one (as the first response could
not be a switch), to get a measure of switching independent of the number of responses
generated by the participant. A higher score indicates more frequent shifts between cue words.

Mean number of clusters. We defined a cluster as two or more consecutive responses that were
assigned to the same cue word. For each trial, we counted the total number of clusters. We
divided this by the total number of responses minus one (as the first response could not belong
to a cluster), to get a measure of clustering that was independent of the number of responses
produced by the participant. A higher score reflects a stronger tendency to form clusters of
associated words.

98
Largest cluster size. Cluster size was computed as the number of successive responses assigned
to a cluster (plus one, to account for the first item in the cluster). We defined the largest cluster
size as the maximum cluster size across all clusters generated during a trial.

Average cluster size. The average cluster size was computed as the mean cluster size across all
clusters within a given trial. Higher average cluster sizes indicate a tendency to form larger
clusters.

First switch rank. This reflected the position of the first switch in the response chain,
representing the number of responses related to a single cue before the first switch occurs.
Higher switch ranks suggest that participants tend to stay within the semantic space of the first
cue for longer before switching to a different cue word.

Sequential dependence. The sequential dependence in each trial was calculated as the average
semantic similarity across all consecutive response pairs on a given trial. Higher sequential
dependence scores suggest that participants’ responses were driven by associative chaining.

4.2.5. Analysis Strategy

First, we examined the associations between the various performance measures using
Pearson’s correlations. To answer our first research question, we ran a series of mixed effects
models to examine whether individual differences in generation and evaluation abilities predict
performance on creativity tasks. First, we ran a generalised linear mixed effects model to
examine whether generation and evaluation abilities effect RAT performance. This was a
binomial model as it predicted RAT success at the individual trial level (solved/not solved).
Second, we ran linear mixed effects models to estimate the effect of these abilities on AUT
Originality, AUT Uniqueness and AUT Fluency. All mixed effects model were fit with
maximum likelihood estimation and the bobyqa optimiser (lme4 package; Bates et al., 2015).
Random by-participant and by-trial intercepts were included to account for individual and trial-
level variation. Continuous predictors were standardised before inclusion. Denominator
degrees of freedom were approximated using the Satterthwaite method (Kuznetsova et al.,
2017). Elaboration, an important confound in AUT scoring, was included as a covariate in all
AUT models.
To answer our second research question, we extracted latent factors underlying our
clustering/switching metrics by running a principal components analysis on the clustering and
switching indices derived from the G-RAT. Minimum Average Partial (MAP; Velicer, 1976)
method and Parallel Analysis (Horn, 1965) were used to find the optimum factor solution. Two

99
factors with eigenvalues greater than one were extracted and varimax rotated, and component
scores were used as predictors in the final analyses. Component 1 represented clustering depth,
and Component 2 represented clustering breadth. To understand the effects of semantic
selection and the clustering/switching metrics on RAT accuracy at the single-trial level, we ran
a generalised linear mixed effects model with the following predictors: cluster Depth, cluster
breadth, E-RAT score, sequential dependence, and number of responses generated. To explore
whether there was a trade-off between the benefits of clustering vs. switching moving around
the semantic space, we used a likelihood ratio test (LRT) to investigate whether an interaction
term for cluster depth and cluster breadth improved model fit. We then re-ran the same analyses
but using the original clustering/switching scores (average cluster size, mean clusters) and their
interaction. Finally, we ran linear mixed effect models to examine the effects of the semantic
predictors (cluster size, cluster number, mean responses and E-RAT scores) on AUT
Originality, AUT Uniqueness and AUT Fluency. Again, an LRT was used to compare the fit of
an additive and interaction model.

4.3. Results

4.3.1. Relationships between Cognitive Measures and Creative Performance

Table 1 demonstrates the means and standard deviations of the variables of interest.
Participants correctly solved an average of 3.07 Standard RAT problems (20.47%) and 9.76 E-
RAT problems (48.8%). In the G-RAT, participants’ response chains contained the solution on
an average of 0.15 trials (15%). There was some variation in the number of responses generated
on both the AUT (M = 6.18, SD = 2.47, Range = 2.50-16.50) and the G-RAT (M = 10.01, SD
= 4.08, Range = 5.10-31.80).

100
Table 1: Descriptive Statistics for Cognitive Measures.
Measure Units Mean SD Min Max SE

Standard RAT Total correct 3.07 2.01 0.00 8.00 0.19

(max15)

AUT Fluency No of Responses 6.18 2.47 2.50 16.50 0.24

per item

AUT Unique Uncommonness 0.27 0.24 0.00 1.38 0.02

of responses
AUT Originality AI-based scoring 2.71 0.24 1.84 3.40 0.02
AUT Elaboration No. of words 2.44 1.28 1.00 8.42 0.12
per response

Generation RAT Number of 10.01 4.08 5.10 31.80 0.39

responses per
trial
Generation RAT Proportion 0.15 0.13 0.00 0.50 0.01
Accuracy correct

Evaluation RAT Total correct 9.76 3.00 3.00 19.00 0.29

(max 20)

Creative Behaviours Self-Report 55.38 12.35 29.00 85.00 1.18

Inventory

Number of Clusters Mean number of 0.26 0.04 0.15 0.38 0.00

clusters minus 1

Number of Switches Mean number of 0.55 0.09 0.30 0.75 0.01

switches minus 1

Average Cluster Size Mean Size of 1.85 0.35 1.25 2.98 0.03
Clusters

Max Cluster Size Largest Cluster 2.44 0.71 1.33 5.00 0.07
Size

First Switch Rank Position of First 3.02 0.57 2.20 4.70 0.05
Switch
Sequential Mean sequential 0.19 0.03 0.11 0.30 0.00
Dependence dependence
Note. RAT = Remote Associates Test, AUT = Alternate Uses Task, SD= Standard Deviation, Min = Minimum, Max = Maximum, SE = Standard Error

Pearson’s correlations were used to investigate the relationships between the various
cognitive abilities (see Figure 2). Standard RAT performance was positively related to being
more likely to get to the right answer on the G-RAT and to producing more clusters in the G-

101
RAT; it was also negatively related to the average size of the clusters. Further, measures related
to the quantity of ideas generated were related with one another: AUT fluency was correlated
with the number of responses generated on the G-RAT, as well as to AUT Elaboration (i.e., the
average number of words produced in a response). E-RAT performance, indexing the ability to
appraise suitability and select from candidate responses, was positively related to accuracy on
the G-RAT, but negatively related to AUT Uniqueness. Finally, the AUT metrics showed
patterns of correlations amongst themselves: AUT Uniqueness was positively associated with
AUT Originality and AUT Elaboration, but not AUT Fluency.
The metrics indexing clustering and switching tendencies derived from the G-RAT were
all highly interrelated: finding the right answer on the G-RAT was related to the number of
responses generated; the measures of cluster size were positively correlated with each other;
the mean number of clusters generated was negatively associated with the mean number of
switches from one subcategory to another; and the mean number of switches were negatively
related to measures related to size of the cluster and whether the search exhibited sequential
dependence. Cluster size was also related to sequential dependence, indicating that the
elements within larger clusters tend to be semantically similar to one another.

Figure 2: Pearson correlations between measured abilities. (a) Relationships between

original measures; (b) Relationships between PCA-derived measures of clustering, switching
and G-RAT abilities. Note: RAT = Remote Associates Test; AUT = Alternate Uses Test; G-
RAT = Generation RAT. * = p<.05; ** = p<.01; *** = p<.001

102
4.3.2. Generation and Evaluation as Predictors of Creative Thinking
We used mixed effects regression models to test whether the ability to generate and
evaluate candidate ideas (i.e., G-RAT and E-RAT performance) predicted performance on the
standard RAT and AUT. Effects are shown in Table 2. First, the logistic mixed effects model
indicated that neither G-RAT responses nor E-RAT accuracy significantly predicted Standard
RAT scores. Generation and evaluation were differentially related to AUT measures: in the
linear mixed effects model, evaluation negatively predicted Uniqueness, but not Originality
or Fluency; while generation positively predicted Fluency. Neither ability was related to self-
reported creativity in everyday life. These results suggest that people who are skilled at
generating ideas under multiple cue constraints, as in convergent thinking tasks, are also
more able to produce ideas in a divergent thinking task. However, people with a greater
ability to evaluate ideas for appropriateness tend to produce less original AUT responses.
This indicates that a strong focus on evaluation may inhibit creativity in divergent thinking.

Table 2: Effects of generation and evaluation abilities on creative performance.

Outcome Generation Evaluation
Estimate CI p Estimate CI p
Standard RAT 0.89 0.73 – 1.10 0.298 1.06 0.87 – 1.30 0.555
AUT Uniqueness 0.01 -0.03 – 0.05 0.584 -0.05 -0.09 – -0.00 0.029
AUT Originality 0.01 -0.03 – 0.06 0.550 0.00 -0.04 – 0.05 0.919
AUT Fluency 0.71 0.26 – 1.16 0.002 0.17 -0.28 – 0.62 0.461
Creative Behaviours 2.07 -0.26 – 4.41 0.081 0.37 -1.96 – 2.71 0.752

4.3.3. Principal Components Analysis

Given that our five clustering/switching metrics derived from the G-RAT were all
strongly related, we ran a principal components analysis to explore the underlying structure of
these clustering and switching behaviours. We did not include sequential dependence here, as
we were interested in testing this as an independent predictor. A 2-component solution was
suggested by the Minimum Average Partial (MAP) method (Velicer, 1976) and Parallel
Analysis (Horn, 1965), with both components together accounting for approximately 85% of
the variance. Table 3 shows the components loadings with varimax rotation. The first
component seemed to capture clustering depth, with positive loadings of largest cluster size,
average cluster size, and first switch rank and negative loadings of mean number of switches.
High scores on this factor suggest a tendency to form large clusters, switch later in the response
chain, and switch less overall. Thus, it can be surmised that this component captures the ability

103
to maintain retrieval within a semantic subcategory, representing a strategy focused on fully
exploiting one area of the semantic space, with less movement to other areas.
On the other hand, the second component captures the clustering breadth of search,
with strong positive loadings from the mean number of clusters, and negative loadings of mean
number of switches. This reflects the diversity of the semantic categories explored during
search, with high scores suggesting an inclination to explore a range of different subcategories
within the semantic space, though without continually switching. To sum, the two PCA-derived
factors seem to capture independent aspects of clustering behaviour: their depth and breadth.
We used these component scores in subsequent analyses. Figure 2 demonstrates the correlations
between the two factors and the overall G-RAT performance measures. People scoring high on
cluster depth tended to produce more responses but neither factor was related to the likelihood
of identifying the correct solution.

Table 3: Exploratory Principal Components Analysis of clustering/switching metrics.

Component 1: Component 2:
Cluster depth Cluster breadth

Average Cluster Size .94 -.07

Largest Cluster Size .93 -.04
First Switch Rank .71 .41
Number of Switches -.72 -.64
Number of Clusters -.07 .98

SS Loadings 2.79 1.54

Proportion Variance 0.56 0.31
Cumulative Variance 0.56 0.87
Note: Factor loadings > 0.4 are shown in bold.

4.3.4. Semantic search components as predictors of Standard RAT performance

First, we examined the effects of the PCA-derived search components on standard RAT
performance using logistic mixed effects modelling. This model included G-RAT and E-RAT
performance as predictors, as well as sequential dependence. A likelihood ratio test indicated
that the inclusion of an interaction between the two PCA factors improved model fit, χ2 (1) =
8.23, p<.01. Results of the final model are shown in Table 4. As in the earlier analysis, overall
generation and evaluation abilities, indexed by G-RAT and E-RAT, did not predict standard

104
RAT success. Nor was there an effect of sequential dependence, suggesting that the propensity
to chain responses associatively is unrelated to how effectively people can solve RAT
problems. There was, however, a main effect of cluster breadth. People who generated many
clusters of candidate solutions in the G-RAT were more likely to solve standard RAT problems.
Furthermore, this effect interacted with the cluster depth factor. This interaction is plotted in
Figure 3a. People who demonstrated high cluster breadth (shown in blue) were most likely to
solve standard RAT problems if their clusters were small in size (i.e., low cluster depth). In
contrast, people with low cluster breadth (shown in red) were more likely to succeed if their
clusters were larger (i.e., high cluster depth). These results suggest that multiple search
strategies can lead to success in RAT problems. Either one can repeatedly visit different areas
within semantic space (high breadth), in which case it is better to keep the number of concepts
retrieved at each visit smaller (low depth). Alternatively, one can focus one’s efforts in a small
number of areas of semantic space (low breadth) in which case it is advantageous to generate
as many concepts as possible when dwelling in each area (high depth).

Table 4: Effects of the PCA components on Standard RAT performance.

OR CI p
G-RAT 0.98 0.78 – 1.22 0.826
E-RAT 1.12 0.93 – 1.35 0.236
Sequential Dependence 0.92 0.74 – 1.14 0.439
Cluster Depth 0.87 0.68 – 1.10 0.241
Cluster Breadth 1.23 1.01 – 1.49 0.041
Cluster Depth x Cluster 0.72 0.58 – 0.90 0.003
Breadth

105
Figure 3: Predictors of Standard RAT Performance. (a) Predicted effects of PCA-derived
search components on Standard RAT performance; (b) Predicted effects of original search
measures on Standard RAT performance.

To further corroborate these findings, we re-ran the multilevel logistic regression

analysis replacing the PCA factor scores with the original measures that contributed most
strongly to each component, namely Average Cluster Size and Number of Clusters (see Table
5). The likelihood ratio test again indicated that an interaction between these variables was a
better fit for the data, χ2 (1) = 15.29, p<.001. Again, the relationship between the average size
of the clusters and RAT accuracy was moderated by the number of clusters formed, as
indicated by a significant interaction (see Figure 3b) (OR = 0.75, 95% CI [0.65 – 0.87], p
<.001. Larger cluster sizes decreased the odds of success on the RAT when individuals
formed more clusters; while larger cluster sizes predicted greater success in individuals who
formed fewer clusters. Finally, there were also main effects of cluster size and number of
clusters: overall, larger cluster sizes decreased the odds of success, while a greater number of

106
clusters led to higher odds of correctly answering RAT questions (OR = 1.19, 95% CI [1.04-
1.70], p = .002).

Table 5: Effects of mean clustering and switching on Standard RAT performance.

OR CI p
G-RAT Responses 0.98 0.85 – 1.14 0.78
E-RAT Accuracy 1.11 0.97 – 1.26 0.14
Sequential Dependence 0.91 0.89 – 1.06 0.24
Average Cluster Size 0.83 0.71 – 0.97 0.018
Number of Clusters 1.20 1.05 – 1.37 0.010
Average Cluster Size × 0.75 0.65 – 0.86 <.001
Number of Clusters

4.3.5. Semantic search components as predictors of AUT performance

Linear mixed effects models were used to examine the effects of the PCA-derived
components on AUT metrics, using a similar approach to the standard RAT analysis. For AUT
uniqueness, the model with an interaction term was a better fit for the data, χ2 (1) = 7.06,
p<.001. Results of the final model are in Table 6. The effects of cluster depth on AUT
Uniqueness depended on cluster breadth, as illustrated by the significant interaction term (β =
-.06, 95% CI [-0.11– -0.02], p = .010). Specifically, when individuals formed more clusters,
the effect of cluster size on uniqueness decreased, implying that the tendency to form larger
clusters may be less useful when individuals form multiple clusters. This interaction is
illustrated in Figure 4a. People who demonstrated low cluster breadth (seen in red) were more
likely to generate unique responses if they fully exploited the area of semantic space (i.e.,
high cluster depth). Conversely, people who demonstrated high cluster breadth (seen in blue)
tended to produce more unique responses if their clusters were smaller (i.e., low cluster
depth). This interplay between cluster depth and breadth is similar to the interaction effect
seen in the earlier RAT models. This suggests that the tradeoff between these two search
strategies is not only beneficial to convergent thinking, but may generalise to other forms of
creative thought as well.
Finally, replicating the earlier analyses, there was a significant main effect of E-RAT
(β = -.04, 95% CI [-0.08– -0.00], p = .038), indicating that evaluation abilities were inversely
related to uniqueness. AUT Elaboration (i.e., the number of words in a response, β = .05, 95%
CI [0.01– 0.09], p = .026) was also positively related to uniqueness.

107
 Table 6: Effects of PCA-derived components on AUT Metrics.
Effect AUT Unique AUT Originality AUT Fluency
B CI p B CI p B CI p
Elaboration 0.04 0.00 – 0.08 0.026 0.07 0.02 – 0.12 0.015 -0.53 -0.98 – -0.08 0.021
G-RAT Responses -0.01 -0.05 – 0.04 0.695 0.01 -0.05 – 0.07 0.667 0.84 0.32 – 1.36 0.001
E-RAT Accuracy -0.04 -0.08 – -0.00 0.039 0.01 -0.05 – 0.06 0.888 0.17 -0.27 – 0.62 0.443
Sequential Dependence -0.02 -0.07 – 0.02 0.313 -0.01 -0.07 – 0.04 0.381 -0.03 -0.52– 0.46 0.901
Cluster Depth 0.03 -0.02 – 0.08 0.261 0.00 -0.06 – 0.07 0.883 -0.20 -0.75 – 0.36 0.482
Cluster Breadth 0.00 -0.04 – 0.04 0.884 0.02 -0.03 – 0.07 0.216 -0.07 -0.52 – 0.39 0.772
Cluster Depth × -0.06 -0.11 – -0.02 0.010
Cluster Breadth

When repeating the analysis using the original measures of average cluster size and
number of clusters in place of PCA factor scores, the interaction model was again a better fit
for the data, χ2 (1) = 5.46, p<.01. Here, we found similar effects: a significant interaction,
indicating that average cluster size was moderated by number of clusters (β = -.05, 95% CI [-
.010 – -.01], p = .023); and conditional main effects of evaluation (β = -.04, 95% CI [-.08 - -
.01], p = .043) and elaboration (β = .05, 95% CI [.01 - .09], p = .019) (see Figure 4b).
Results of linear mixed models predicting AUT Originality and AUT Fluency are also
shown in Table 6. The interaction between PCA factors did not improve fit of neither the
Originality (χ2 (1) = .52, p = 0.46) nor the Fluency model (χ2 (1) = .10, p = 0.74), so this term
was not included. The model replicated effects of elaboration and G-RAT responses (β = .85,
95% CI [0.35– 1.34], p = .028) found earlier, but no effects of our PCA-derived components
of cluster depth and cluster breadth were found.

108
Figure 4: Predictors of AUT Performance. (a) Predicted effects of PCA-derived search
components on AUT Uniqueness; (b Predicted effects of original search measures on AUT
Uniqueness

4.4. Discussion
The aims of the current study were two-fold: first, to understand how the ability to
generate and evaluate candidate responses to multiply constrained problems is related to
performance on standard tasks of creativity; and second, to investigate whether individual
differences in search strategies employed during this problem-solving paradigm can predict
creative ability. To this end, we deconstructed the standard RAT into two measures indexing
generation and evaluation of creative ideas. While neither generation nor evaluation was related
to standard RAT performance, we found that the ability to generate multiple ideas predicted
fluency on the AUT, and the ability to evaluate candidate responses was negatively related to
AUT uniqueness. To gain a more nuanced understanding of the retrieval strategies employed

109
when searching through semantic space, we extracted various metrics of clustering and
switching behaviours from the generation task. Two components indexing clustering depth and
clustering breadth were identified using principal components analysis and used in subsequent
regression analyses. We found that an interaction between these two components indexing
clustering breadth and depth predicted the odds of success on the RAT as well as AUT
uniqueness.

4.4.1. How do generation and evaluation predict creative performance?

First, we found that the ability to generate candidate responses to a multiply constrained
problem is related to AUT Fluency, in line with previous work linking the strategic retrieval of
items from memory (broad retrieval ability; Carroll, 1993) to both the quantity and quality of
creative ideas (e.g., Beaty et al., 2014; Forthmann, Jendryczko, et al., 2019; Miroshnik et al.,
2023; Silvia et al., 2013). Broad retrieval ability is commonly assessed with verbal fluency
tasks wherein participants generate responses based on a single constraint. Our study adds to
these extant findings, demonstrating that the ability to generate responses while considering
multiple cues is also related to the quantity of ideas produced on the AUT. Together, these
results suggest a common generative ability may allow the individual to effectively retrieve
task relevant information and access more remote concepts. However, it is important to note
that we only found that generation predicted the quantity of creative ideas, but not their
originality or uniqueness. These results might be explained by recent meta-analytic evidence
indicating that, while broad retrieval ability is related to divergent thinking, this relationship is
much stronger for measures of AUT fluency rather than quality (Miroshnik et al., 2023; Weiss
et al., 2024).
Contrary to previous work demonstrating that performance on semantic fluency tasks
predicts both divergent and convergent thinking (Lee & Therriault, 2013), we did not find that
the ability to generate multiple responses was related to RAT performance. A potential
explanation for this may come from the multiply constrained nature of the RAT. Unlike
semantic fluency tasks where individuals generate responses within a single category, to
successfully solve the RAT, one must establish a point of overlap between the semantic
neighbourhoods of all three cue words. Thus, despite our generation task being structurally
similar to the RAT, retrieval of a large number of responses may itself not be sufficient to
guarantee success, as this may not capture how efficient the individual is at arriving at the
semantic intersection of the cues. Instead, as we discuss in the next section, our results suggest

110
that the specific cognitive strategies used during semantic retrieval may be a better predictor of
success.
On the other hand, performance on the evaluation task was negatively related to AUT
uniqueness, but not to fluency or originality. This finding is somewhat surprising, given meta-
analytic evidence establishing a small but positive relationship between divergent thinking and
evaluative skill (Guo et al., 2022). However, these previous studies asked participants to
evaluate ideas produced on divergent thinking tasks on novelty and effectiveness criteria; while
our task required them to evaluate associations and select a ‘target’ from a response array,
which may involve more analytic skills. Indeed, when evaluating alternate uses for an object,
participants may be relying episodic rather than semantic retrieval mechanisms (Orwig, Beaty,
et al., 2024). In contrast, our evaluation task was more semantically focused, requiring
participants to evaluate associative links between the cue words and response options.
Moreover, there is some evidence to suggest that the source of ideas matter, with studies
indicating that more creative people are both worse at evaluating the ideas produced by others
(Grohman et al., 2006), and tend to underestimate the creative quality of the responses (Guo et
al., 2019). In the context of our task, it is possible that it may be harder to evaluate pre-
determined associative links between the three cues and each of the response options.
Our results also differ from previous findings establishing that the ability to select
among competing semantic options is a positive predictor of AUT uniqueness (Patel et al.,
submitted). This discrepancy in findings may lie in the task demands: in Patel et al.’s study,
participants had to match a probe word with a target based on a specific property (e.g.,
‘colour’), requiring them to focus on specific feature-related associations of the item, while
suppressing competition from more dominant associations. In contrast, our evaluation task was
much more cognitively demanding - in addition to the standard RAT stimuli of three cue words,
participants were given three potential response options and asked to select the most
appropriate one. To solve the problem, participants had to consider each option against the
intersection of semantic associations between the three cues whilst also engaging in
competition resolution and inhibition. Individual differences in semantic knowledge and
working memory capacity may be important confounds to consider here: larger stores of
conceptual knowledge may provide a broader search space, while high working memory
capacity may be needed to hold all six words in mind while evaluating each option for
appropriateness. (Ellis et al., 2021; Ellis & Brewer, 2018). While these cognitive abilities were
not measured in the current study, we encourage future work to examine the effects of
evaluation on convergent thinking while taking these into account.

111
4.4.2 How do retrieval strategies influence creative performance?
We also aimed to gain a more nuanced understanding of the retrieval strategies that
individuals use during semantic search with multiple cue constraints. Our generation task
captured two components of search behaviour: clustering depth, a tendency to maximally
exploit a cluster of concepts relating to a single cue word; and clustering breadth, the formation
of many different clusters during the search for a solution. Forming a number of different
clusters was related to better performance on the RAT, as this may allow participants to sample
the semantic space related to each of the three cue words more broadly. In contrast, spending
large amounts of time in one part of the semantic space was overall related to worse
performance on the RAT, as becoming stuck in one semantic cluster may cause perseveration
and prevent the consideration of other candidate ideas. Interestingly, we found a complex
interplay between these components, suggesting multiple possible routes to success on the
RAT. For individuals who formed many clusters, it was less advantageous to stay in any one
cluster for a long time; whereas for people who formed few clusters, a more thorough
exploitation of the search space related to a single cue was beneficial. Thus, our findings
highlight the importance of considering both the size and quantity of clusters formed when
individuals search through their semantic space. Exploring a broad range of associations and
shifting between conceptual clusters may help to mitigate the perservation encountered when
over-exploiting a single cluster – but, on the other hand, the shallow exploration of multiple
different clusters may not be helpful either. Furthermore, this trade-off between cluster depth
and cluster breadth may align with the balance between exploitation and exploration advocated
in optimal foraging models (Hills et al., 2012), emphasising the need to switch before the
retrieval rate in the current semantic cluster falls below a certain threshold.
Our results align with the anti-clustering behaviour seen in Davelaar (2015), with more
optimal search involving frequent transitions between cue patches to maximize the difference
in memory activation between the target and distractors. We found that forming a greater
number of clusters was positively related to solving RAT problems, and in people who did so,
forming very large clusters was detrimental to performance. Avoiding deep exploitation of
specific areas of the semantic space may be particularly advantageous to performance on the
RAT, as this would allow for the consideration of the multiple cues involved and the
intersection of their semantic spaces. In contrast, Smith et al. (2013) found that, while adjacent
responses were more likely to be semantically similar and belong to the same cluster, there
were sequential dependencies within and across cluster breaks, suggesting a direct dependence

112
on prior responses rather than a purely cluster-based search. Conversely, our results did not
support this interpretation: while there was some sequential dependence between responses,
this was not predictive of success on the RAT.
Our findings suggest that these search components relate to individual differences in
retrieval dynamics. The results broadly support the findings of Ovando-Tellez et al. (2022).
These researchers used an ambiguous word-fluency task where people had to name all the
words associated with a polysemous cue word. They found that the ability to generate different
meanings and to switch between meanings were both correlated with convergent thinking
performance. Similarly, in our work, it was beneficial to adopt a search strategy that allowed
the formation of multiple clusters to sample the search space more widely, and to avoid being
stuck within a single cluster. Further, these search components can be partially mapped on to
the flexibility and persistence pathways of creativity (Nijstad et al., 2010). While clustering
represents persistence (i.e., the deep exploration of an area of the semantic space with an aim
to maximize retrieval), switching reflects a broader, more controlled search over the whole of
the semantic space. While deep exploitation of a cluster may prompt control processes that lead
away from stereotyped ideas (Beaty & Silvia, 2012), the danger lies in being mired in one
single region of the semantic space. On the other hand, while switching to other areas of
semantic space may counteract perseveration and allow for the discovery of new conceptual
inputs, continuous switching may produce interference and lead to only a shallow exploration
of each area. Importantly, recent work suggests that these two pathways may be simultaneously
operating during the idea generation process (Peterson & Pattie, 2024), which supports the idea
that effective semantic search involves a dynamic interplay between cluster depth and cluster
breadth, as observed in our study.
The complex interplay between cluster size and the number of clusters was also related
to AUT uniqueness, suggesting that when individuals formed more clusters, the influence of
cluster size on uniqueness diminished. This broadly suggests that, for AUT uniqueness, when
searching through memory, the tendency to form larger clusters may be less advantageous when
accessing multiple semantic areas, compared to forming a few large clusters. This roughly
aligns with previous work by Ovando-Tellez et al. (2022) who found that AUT performance
was predicted by a clustering component, indexed by the number of responses, position of the
first switch, and average cluster size on the ambiguous word-fluency task. However, it is
important to note that, in their study, clustering was only related to AUT metrics that tapped
into fluency, suggesting that the correlation between the two constructs was driven by an aspect
of fluent idea generation. In contrast, studies examining clustering and switching behaviour

113
within an AUT task itself suggest contradictory results. First, Hass (2017), using response time
distributions, suggested that the AUT evoked fewer temporal and semantic clusters overall,
with people tending to retrieve close associates of the prompt before moving to more remote
responses related to the overall global context. In contrast, Mastria et al. (2021), using a coding
scheme to categorise AUT responses as ‘category stay’ and ‘category switch’ , found that switch
responses were subjectively rated as being more creative. These conflicting results suggest that
individual differences in the utilization of these search components may vary across task
contexts and methods of computing cluster/switch measures.
While some studies report similar patterns of clustering and switching in visual and
verbal fluency tasks (Hills et al., 2008, 2010), it is important to highlight that both these tasks
involve similar constraints (i.e., participants are given a limited amount of time to generate as
many items as possible from a given category). Indeed, the constraints posed by our two tasks
are quite different: the RAT requires the generation of novel links between multiple disparate
concepts, whereas the AUT involves producing responses that adhere to the two constraints of
novelty and effectiveness. In fact, engagement of cognitive strategies also changes with the
level of difficulty of the RAT problem (Sio et al., 2022; Valba et al., 2021). Taken together, this
suggests that differences in task characteristics and cue dimensions may inherently necessitate
different search strategies. Moreover, recent work suggests that clustering and switching may
not be individual trait-level behaviours at all, but instead may be adaptive cognitive strategies
used flexibly to match contextual demands (Mekern et al., 2019). More work is needed to
determine the extent to which search strategies remain consistent within individuals or are a
function of task demands.
Finally, it is important to acknowledge recent critiques regarding the modified version
of the RAT used by us and by Davelaar (2015) and Smith et al. (2013). While the task was
designed to elicit search processes in real time, Howard et al. (2020) argue that the design of
the task may change the nature of the search process. Writing down each candidate idea may
prompt participants to use these candidates as cues for successive responses, evoking the
sequential dependence seen in the responses. As the solution to multiply constrained problems
lies within the intersection of the cues, Howard et al. (2020) suggest that individuals may
naturally use parallel search processes (i.e., considering the three cues simultaneously) that are
related to more a rapid, implicit search. The more explicit, deliberative nature of our task may
inadvertently encourage serial processing, suggesting that our results could reflect the specific
constraints and influences of the task design. Future studies examining semantic search with

114
multiple cue constraints should take these inherent tendencies into consideration when
designing tasks to better capture these processes.
In conclusion, our results demonstrate that the ability to generate responses is
generalisable across tasks with varying cue constraints, supporting the idea that a broad
retrieval ability contributes to the quantity of creative ideas produced. Additionally, our results
suggests that the ability to evaluate candidate responses in a RAT context may not align with
evaluative skills needed for divergent thinking tasks. Finally, our findings highlight the
complexity of the search processes involved in creative thinking. Importantly, our findings
underscore the importance of considering both the size and quantity of clusters formed when
individuals search through their semantic space. The nuanced interplay between cluster depth
and breadth indicates multiple possible routes to success on the RAT and the AUT, suggesting
that effective semantic search involves a balance between exploitation and exploration.

115
 Chapter 5

Balancing Act: A Trade-off Between Coherence and Creativity in

Spontaneous Speech

Abstract

The ability communicate effectively involves a balance between generating novel,

engaging content and maintaining a coherent narrative. While both creativity and coherence
form an important part in making a speaker interesting and comprehensible, the mechanisms
underlying this balance remain unexplored. The aim of the current study was to investigate the
relationship between coherence and creativity in spontaneous speech, with a specific focus on
the interaction among three key neural networks: the Default Mode Network, Multiple Demand
Network and the Semantic Control Network. We carried out two studies examining this
relationship at the behavioural and neural levels. Study 1 aggregated a large set of verbal
responses to topic prompts from previously collected datasets examining spontaneous speech
in younger and older adults (e.g., Hoffman et al., 2018; Hoffman, 2019; Morales et al., 2022;
Patel et al., 2023). Using computational linguistic techniques, we measured global coherence
(indexing the degree of connectedness to the main topic) and Divergent Semantic Integration
(reflecting the diversity of ideas incorporated in the narrative). Coherence and divergence in
speech were negatively correlated, suggesting a trade-off between maintaining narrative
structure and incorporating creative elements. To investigate how large-scale brain networks
support coherence and creativity in naturalistic speech, Study 2 analysed a subset of responses
from a neuroimaging study (Morales et al., 2022; Patel et al., 2023). We employed Independent
Components Analysis on this functional imaging data to extract our networks of interest.
Network activation analyses indicated that higher global coherence was associated with greater
activation in the Multiple Demand Network, emphasising its role in organising and sustaining
logical flow in spontaneous speech production. In contrast, functional connectivity analyses
demonstrated that higher DSI was related to a coupling between default and executive
networks, suggesting that creative speech relies on a dynamic interplay between associative
and executive processes. Interestingly, the Semantic Control Network was not significantly
involved in either coherence or creativity, suggesting its role may be more context-dependent
than previously understood. These results provide new insights into the cognitive and neural
processes underpinning spontaneous speech production, highlighting the complex interaction

116
between different brain networks in managing competing demands of being coherent but
creative.

117
5.1. Introduction
What makes a speaker interesting? Leaving aside the razzle dazzle of gestures,
intonation and other non-verbal communication, an interesting speaker distinguishes
themselves by delivering content that goes beyond the ordinary. These speakers bring new ideas
and provide unconventional perspectives to a given topic, two attributes that are deeply
entwined with creativity. To explore these different angles on a topic, speakers may be relying
on their divergent thinking, i.e., the ability to generate multiple different ideas from a single
starting point (Guilford, 1967). However, to be most effective, they must also ensure that their
communication is coherent, i.e., that these disparate ideas are organized logically and weaved
into a structured and well-connected narrative (Glosser & Deser, 1992). Indeed, a
conversational cue may invoke a range of semantic information, and it is up to the skilled
speaker to select the subset of information that may be most relevant - and, hopefully,
interesting. Recent advances in network neuroscience implicated the Default Mode Network,
the Executive Control Network, and the Semantic Control Network in both creative processes
(Luchini & Beaty, 2023) and the maintenance of coherent speech (Hoffman, 2019; Morales et
al., 2022). However, the interplay between these networks and their specific contributions to
creativity and coherence in spontaneous narrative speech production remain largely
unexplored. To understand the neurocognitive mechanisms that make a speaker engaging and
effective, the present study will explore the relationship between creativity and coherence when
individuals generate extended passages of speech.
Creative thought involves a complex interplay between spontaneous, bottom-up
associative processes and top-down, controlled executive processes (Kleinmintz et al., 2019).
Semantic memory, our stores of knowledge about people, places, and things, plays a central
role in the idea generation stage of this process (Abraham, 2014; Abraham & Bubic, 2015).
Associative theories suggest that this knowledge is stored as a network of interconnected
concepts of varying associative strengths, with the spontaneous spread of activation across this
network resulting in the combination of more remote elements (Beaty & Kenett, 2023; Kenett
& Faust, 2019; Mednick, 1962). Under this framework, the emphasis is on the individual
differences in the structure of semantic memory: creative individuals have many weak
associations between concepts, allowing them to bypass stereotypical connections and
integrate diverse elements more effectively. Indeed, creativity has been linked to network
properties that support the facilitation of remote associations: more creative individuals possess

118
flexible, densely connected networks with shorter path lengths between concepts, enabling
activation to spread more efficiently across the network (Kenett, 2018; Kenett et al., 2014;
Kenett, Medaglia, et al., 2018; Kenett & Austerweil, 2016; Kenett & Faust, 2019).
While novelty is an important aspect of creativity, ideas must also be appropriate to the
constraints of the task (Runco & Jaeger, 2012). To achieve this balance, during the idea
evaluation phase, individuals assess the quality of these ideas by monitoring, evaluating, and
selecting them according to the task goals (Kleinmintz et al., 2019). Several top-down
processes exert cognitive control over creative processes, including domain-general executive
functions like working memory, shifting and inhibition (Benedek, Könen, et al., 2012; Cheng
et al., 2016; Palmiero et al., 2022; Zabelina et al., 2012); fluid and crystallized intelligence
(Cho et al., 2010; Forthmann, Jendryczko, et al., 2019; Nusbaum & Silvia, 2011); and broad
retrieval ability (Beaty & Silvia, 2012; Forthmann, Jendryczko, et al., 2019; Miroshnik et al.,
2023). These domain-general control processes facilitate creative thinking by filtering out
irrelevant distractors; maintaining and manipulating multiple pieces of information in working
memory; and enabling flexible shifting between ideas. Taken together, the literature suggests
that creative thinking arises from a dynamic interaction between these spontaneous and
controlled processes: ideas are generated through associative processes acting on semantic and
episodic memory, and iteratively evaluated for their adherence to the task requirements (Beaty,
Silvia, et al., 2014; Kleinmintz et al., 2019).
This dynamic interaction between spontaneous and controlled processes is mirrored in
brain activity: creative cognition is mediated by an interaction between brain networks
supporting idea generation and goal-directed evaluation (Beaty et al., 2019; Luchini & Beaty,
2023). Creativity neuroscience research has seen a shift from classical perspectives with a focus
on localised brain activity to a network approach, exploring the functional connectivity
between brain networks that are spatially distributed but show correlated patterns of activity in
relation to tasks and at rest. Under this approach, resting state functional connectivity (RSFC)
and task-based functional Magnetic Resonance Imaging (fMRI) studies have identified two
key networks that dynamically interact to produce novel and appropriate ideas: the Default
Mode Network (DMN) and the Executive Control Network (ECN) (Adnan, Beaty, Lam, et al.,
2019; Adnan, Beaty, Silvia, et al., 2019; Beaty et al., 2015; Shi et al., 2018; Zhu et al., 2017).
The DMN consists of a distributed set of regions across the frontal, temporal and parietal
cortices, including medial prefrontal cortex (mPFC), posterior cingulate cortex (PCC),
precuneus, temporoparietal junction, and the bilateral inferior parietal lobes (IPL; Smallwood
et al., 2021). Typically, the DMN shows reduced activation in relation to externally focused

119
tasks, but increases its activation when individuals are engaged in spontaneous and self-
generated thought (Andrews-Hanna, 2012; Andrews‐Hanna et al., 2014). Consequently, DMN
activity has been associated with a variety of internally directed cognitive processes, including
mind-wandering (Christoff et al., 2009, 2016; Fox et al., 2015), autobiographical retrieval
(Spreng et al., 2009), mental simulation (Hassabis & Maguire, 2007), and episodic future
thinking (Cona et al., 2023; Schacter et al., 2012). The DMN has been reported to be active
during divergent thinking in several studies (Beaty, Benedek, et al., 2014; Benedek et al., 2014;
Ellamil et al., 2012; Fink et al., 2018; Gonen-Yaacovi et al., 2013; Shofty et al., 2022; Takeuchi
et al., 2012; Wei et al., 2014). With respect to creativity, the DMN is thought to be involved in
the idea generation phase, integrating self-referential, episodic, and semantic information to
create novel ideas; while the ECN is involved during idea evaluation (Ellamil et al., 2012).
The ECN, consisting primarily of regions in the dorsolateral prefrontal cortex, anterior
cingulate cortex and inferior parietal cortex, is engaged during externally directed cognitive
tasks that require cognitive control (Seeley et al., 2007). This distributed network of regions,
also referred to as the ‘Multiple Demand Network’ (MDN) or ‘Fronto-Parietal Control
Network’ (FCPN), is responsible for top-down control, including cognitive processes such as
working memory, flexibility, and response inhibition (Niendam et al., 2012). Thus, the ECN
may be crucial during the idea evaluation phase of creativity, to assess whether candidate ideas
meet task criteria and to modify them accordingly (Gonen-Yaacovi et al., 2013). Importantly,
creative cognition requires a combination of spontaneous idea generation (mediated by the
DMN) and goal directed processing (mediated by the ECN). The two networks generally share
an antagonistic relationship, with one network deactivating when the other activates (Fox et
al., 2005). However, the networks tend to cooperate during tasks involving internally directed
processes, including episodic future planning and constructive daydreaming (Andrews‐Hanna
et al., 2014; Christoff et al., 2009; Gerlach et al., 2014; McMillan et al., 2013; Spreng et al.,
2010). Similarly, DMN-ECN coupling has been consistently associated with creativity across
a range of tasks, with the DMN contributing to the generation of potential ideas, while the ECN
exerts top-down control to monitor progress in line with task related goals (Beaty et al., 2015,
2016; Green et al., 2015; Liu et al., 2015; Mayseless et al., 2014; Zhu et al., 2017). Indeed,
stronger and more efficient connectivity between these regions is associated with individual
differences in creativity and better performance on creativity tasks (Beaty et al., 2015; Beaty,
Kenett, et al., 2018; Green et al., 2015). Moreover, there is increased coupling through the
creative process, suggesting that creativity benefits from flexible exercise of cognitive control
mechanisms (Zabelina & Robinson, 2010).

120
 Aside from the ECN and domain-general executive functions, semantic control
mechanisms may also have more specific roles in regulating the search, manipulation, and use
of task relevant information in pursuit of creative goals (Cogdell‐Brooke et al., 2020; Krieger-
Redwood et al., 2023). Two semantic control mechanisms have been proposed: first, controlled
retrieval mechanisms are engaged when spontaneous activation does not yield task-appropriate
semantic information (Badre et al., 2005; Badre & Wagner, 2007). Individuals use goal-directed
processes to search for more relevant information, inhibiting dominant associations and
promoting the retrieval of unconventional ideas. Second, semantic selection mechanisms
resolve competition between ideas during the evaluation stage by assessing and selecting the
most suitable ones. The Semantic Control Network (SCN) has been shown to contribute to
creativity over and above the DMN and ECN (e.g., Krieger-Redwood et al., 2023; Rastelli et
al., 2024). The SCN, including the left inferior frontal gyrus, posterior middle temporal gyrus
and dorsomedial PFC (Jackson, 2021; Noonan et al., 2013), plays a crucial role in the flexible
and controlled retrieval of less dominant semantic information, which is essential for creative
thinking. The IFG, and more specifically BA45, has been associated with the semantic selection
across a range of tasks, showing greater activation for tasks involving competition from
irrelevant semantic associations or between possible response options (Badre & Wagner, 2007;
Thompson-Schill et al., 1997; Whitney et al., 2011b). Key areas of the SCN, specifically the
left IFG and dorsomedial PFC, are recruited when generating creative links between concepts
(Krieger-Redwood et al., 2023) and during the production of creative story narratives (Rastelli
et al., 2024), especially when the level of semantic control demands are manipulated. Given
the spatial adjacency of the SCN to the DMN and ECN, this may allow for flexible semantic
retrieval, i.e., the manipulation of semantic knowledge to fit task relevant goals (Chiou et al.,
2023b; Wang et al., 2020). The position of the SCN may allow it to leverage interactions
between these networks, facilitating the generation of novel and creative ideas by directing
attention toward unusual and non-dominant features and associations of concepts.
Building on the insights from neuroimaging studies, it is imperative to consider how
creative thinking is commonly assessed and measured under task-based approaches. Divergent
thinking, a widely used indicator of creative potential, is often operationalised through open-
ended tasks that measure an individual’s ability to generate a variety of ideas that are novel and
task-appropriate. While effective at evaluating creativity at the word and sentence level
(Weinstein et al., 2022), these traditional divergent thinking tasks may not capture the
complexity of creative thinking as it occurs in real-world contexts. Everyday creative thinking
involves more than simply generating individual ideas. It encompasses the ability to develop

121
these ideas into coherent narratives, solve complex problems, and adapt to diverse situations
(Ilha Villanova & Pina e Cunha, 2021). However, work examining more complex creative
outcomes has been limited by the reliance on subjective scoring methods such as consensual
assessment techniques and rubric-based scoring, that are time consuming, labour intensive and
may require expert judges. To address these limitations, advancements in computational
linguistic models have enabled automated methods of scoring longer-form responses, which
show high correlations with human ratings (Orwig et al., 2021). These methods have been used
to assess creativity in more complex domains, such as creative writing (D’Souza, 2021; Fan et
al., 2023; Zedelius et al., 2019), academic writing (Bower et al., 2023), story generation
(Rastelli et al., 2024), and poetry composition (He et al., 2022). Despite these advancements,
one area that remains unexplored is spontaneous narrative speech, an important aspect of real-
time communication that provides a rich avenue for exploring everyday creativity in an
ecologically valid manner.
In the context of narrative speech, being creative may involve generating more
divergent content, i.e., ideas that are more distantly connected to the topic under discussion.
Additionally, it is crucial for the speaker to combine and organize these ideas in a coherent and
engaging manner, ensuring that the overall speech remains understandable. These dual
constraints are supported by findings in the domain of creative writing. Human-based scoring
of story originality is related to the global semantic distance between the words in a story,
reflective of the richness and divergence of the content (Fan et al., 2023). At the same time, the
global cohesion of the story - the degree to which the story ending was related to the context -
is related to human ratings of story rationality, emphasising the importance of building
meaningful connections between ideas and to the main story arc. Importantly, higher global
cohesion was associated with lower originality, suggesting a trade-off where maintaining a
coherent and rational structure may limit the inclusion of novel or unconventional ideas. This
observed trade-off between coherence and divergence in creative writing underscores the
necessity to investigate these dynamics in spontaneous speech. Understanding how individuals
balance coherence and creativity in real-time communication is crucial, as it directly impacts
the effectiveness of everyday interactions.
Maintaining coherence during speech involves the generation of topic-relevant ideas as
well as the monitoring and selection of content to stay on topic. More coherent discourse
successfully navigates these challenges, resulting in a series of well-connected statements that
relate meaningfully to each other and the main topic (Glosser & Deser, 1992). Coherence can
be achieved at multiple levels: globally, by relating each utterance to the overall topic; and

122
locally, by ensuring logical continuity between adjoining sentences. When generating narrative
speech, speakers use situation models, i.e., multimodal representations of events and situations,
to organise their thoughts and plan utterances. Speakers need to construct and use these
situation models in real time, emphasising the key role of executive functioning in planning
and maintaining coherence while speaking (Barker et al., 2017; Kintz et al., 2016; Marini &
Andreetta, 2016). In line with this, improved speech coherence is related to improved cognitive
control in older adults (Hoffman, Loginova, et al., 2018; Kintz et al., 2016; Wright et al., 2014).
Additionally, recent behavioural work highlights the role of semantic control processes in
speech coherence, over and above domain-general executive abilities (Hoffman, 2018).
Specifically, semantic selection abilities positively predicted speech coherence, indicating that
the ability to select a subset of topic-relevant information and suppress other aspects of
knowledge is crucial to coherent communication. In summary, due to the central role of
situation models and control mechanisms (and more specifically, semantic control) in
generating coherent speech, one might expect that the three networks discussed earlier – the
DMN, ECN and SCN - to contribute to coherence.
The DMN plays a critical role in the construction of situation models that represent
different contextual elements and their interrelationships (Yeshurun et al., 2021). Less coherent
discourse places more demands on this system: DMN activation is higher during the production
and comprehension of less coherent naturalistic speech (Morales et al., 2022). This increased
DMN activation may reflect the regular updating and reconfiguring of mental models required
when discourse lacks coherence, suggesting that the DMN is actively engaged in integrating
diverse pieces of information and adapting the narrative as it unfolds.
In addition to the DMN, key areas of the SCN have also been implicated in maintaining
coherence. When older adults produced passages of extended speech, more coherent speech
was associated with increased activation in areas of the prefrontal cortex, namely the bilateral
IFG (BA 45), and the rostro lateral PFC (RLPFC; BA 10; Hoffman 2019). This suggests that
while the DMN integrates higher-level semantic information to form coherent narratives, the
SCN controls and regulates the flow of this information. The SCN’s role may lie in selecting
relevant information and inhibiting irrelevant details to maintain a coherent structure. Listening
to less coherent discourse may additionally tax the SCN, as it may need to work harder to
manage incoming information and maintain a sense of coherence. Interestingly, Morales et al.,
(2022) found less coherent speech was related to increased activation in semantic control
regions of younger adults, but only during speech comprehension. This mixed evidence
suggests that the engagement of the SCN in relation to coherent discourse may be dependent

123
on different age and task related factors that influence task difficulty. For example, when
generating coherent speech, older adults may rely more on the SCN due to age-related declines
in general executive functioning. Similarly, when listening to incoherent speech, younger
people may actively engage semantic control processes to sift through disorganised information
and select relevant details to integrate into a coherent mental model.
Finally, the role of the ECN in coherence is still unclear. Some researchers have
proposed that speech production relies on domain-general executive control regions,
particularly those within the dorsolateral PFC (Wise & Geranmayeh, 2016a). However,
Morales et al., (2022) did not find any modulation of activity in the executive-control related
Multiple Demand Network (MDN) in relation to coherence during the production of
spontaneous naturalistic speech. However, the ECN has been associated with global cohesion
in a story generation task, indicating that the ECN may be significant in relation to tasks that
require more structured and deliberate organisation of content.
Thus far, we have put forth the argument that when generating narrative speech,
individuals must balance a desire to be divergent (and therefore maximally interesting) with
the need to maintain coherence. Evidence from network neuroscience suggests that both these
qualities rely on similar underlying neurocognitive networks. On one hand, creativity is
underpinned by a DMN-ECN coupling that allows for an iterative interplay between bottom-
up associative processes and top-down executive control during idea generation and evaluation.
This coupling is supported by the SCN, enabling flexible, goal-directed retrieval and selection
processes. On the other hand, coherence is underpinned by DMN-related activity that
contributes to building and monitoring of situation models. While both executive functions and
semantic selection are related to coherence at a behavioural level, the exact contributions of the
SCN and ECN to more coherent speech are still unclear. Moreover, the neural correlates of
creativity and coherence in language generation have previously been studied separately. This
is potentially problematic as these qualities are likely to be inter-correlated, with more coherent
speech being less creative and vice-versa. With all this in mind, the present study has two aims:
(1) to explore how creativity and coherence in language production are related at the
behavioural level; (2) to understand how network-level activation and functional connectivity
relates to these two constructs.

124
5.2. Study 1: Methods
In Study 1, we collated a large set of verbal responses to topic prompts from a range of
previous studies. Then, using established methods, we computed measures of coherence and
creativity (divergence) for all responses, to examine the relationship between these constructs.

5.2.1. Participants
Data for the following studies was collected at the University of Edinburgh, School of
Philosophy, Psychology and Language Sciences. All participants were native English speakers,
with no history of neurological or psychiatric illness. The studies received ethical approval
from the Edinburgh Psychology Research Ethics Committee, and informed consent was
collected from all participants.
MD Dataset. This sample consisted of 30 younger adults (MAge= 20.9, range = 19-26), recruited
through the undergraduate participant pool and compensated with course credit, and 23 older
adults (MAge= 73.2, range = 65-89) recruited through the department’s volunteer panel. The
younger sample had an average of 16.3 years of education, while the older sample had 15.0
years of education. These are unpublished data from student projects supervised by PH.
18-ELife Dataset. This sample consisted of 30 younger adults (MAge= 19.3, SDAge = 2.2, range
= 18-30), recruited through the Psychology Department’s undergraduate participant pool, and
30 older adults (MAge= 76.0, SDAge = 8.3, range = 61-91), recruited through the Department’s
volunteer panel. The younger group had an average of 13.8 years of education, while the older
group had 14.3 years of education. Previous analysis of this data, examining speech coherence
in relation to executive and semantic abilities, has been published (see Hoffman et al., 2018).
19-NC Dataset. The sample consisted of 15 older adults (MAge= 77.5, range = 67-92), who had
an average of 15.0 years of education, recruited via the Department’s volunteer panel. This data
has previously been used in a study examining the neural correlates of coherence in speech
production (Hoffman, 2019).
SL19 Dataset. The sample consisted of 25 participants (MAge= 24, SDAge = 4.4, range = 18-35,
21F, 4M) recruited through the University as part of my Master’s dissertation project. This data
has been part of several prior analyses examining neural networks related to coherence
(Morales et al., 2022), the neural correlates related to psycholinguistic properties of naturalistic
discourse (Wu, Morales, et al., 2022) and the neural encoding of meaning during speech
production and comprehension (Patel et al., 2023).
Current sample. Combining the above datasets resulted in a final sample of 153 participants
(102F, 51M), of which there were 85 younger (MAge= 21.08, SDAge = 3.32, min age = 17.98,

125
max age = 35, 65F, 20M) and 68 older (MAge= 75.53, SDAge = 7.75, min age = 61.65, max age
= 91.41, 37F, 31M) adults. A demographic summary for all datasets is presented in Table 1.

Table 1: Demographic summary of datasets.

No of
Age f/m
Prompts
Group n M SD Min Max Count Count
MD 53 43.36 26.25 19 89 35/18 12
18-EL 60 47.53 29 17.98 91.41 37/23 14
19-NC 15 76.91 8.09 65.81 91.41 9/6 20
19-SL 25 23.83 4.29 18 35 21/4 12
Combined 153 47.14 27.83 17.98 91.41 102/51 27

Young Adults 85 21.08 3.32 17.98 35 65/20 26

Older Adults 68 75.53 7.75 61.65 91.41 37/31 20

Note: n = Number of Participant, M = Mean, SD = Standard Deviation, Min.= Minimum, Max = Maximum, f = Female, m = Male.

5.2.2. Design and Procedure

In each of the experiments, participants were given a series of prompts and asked to generate
speech for a period of between 50 and 60s. The prompts were constructed to probe specific
aspects of semantic knowledge (e.g. ‘What would it have been like to live in the Middle Ages?’).
For a full list of prompts used in each individual study, please see Appendix C (Table S7).
MD Dataset. Participants were presented with a prompt on a computer screen, which was
followed by a tone that cued them to start speaking about the topic. They were asked to stop
speaking when they heard the second tone. Participants responded to prompts in three
conditions, each with a different set of goals. In the “Interesting” condition, participants were
instructed to make their responses as interesting, fascinating, and entertaining as possible; in
the “Clear” condition, participants had to keep their responses as clear, focused, and simple as
possible. Finally, participants were given no instructions in the “Baseline” condition. The order
of conditions was counterbalanced between participants. Each participant responded to 12
prompts.
18-ELife Dataset. In this study, speech was elicited under two conditions, dual-task and
speech-only, with seven speech samples elicited in each condition. On speech-only trials,
participants were presented with a written prompt and asked to press a key when they were
ready to begin. They were then asked to speak about the topic until they heard a tone, with each

126
trial lasting 60 seconds. The dual-task trials, where participants produced speech while
completing an attention demanding task, were excluded from the current analysis.
19-NC Dataset. Participants were presented with a series of prompts addressing a range of
semantic knowledge while undergoing fMRI. At the start of each trial, there was a preparation
phase, where the written prompt was presented on screen for eight seconds. When they saw a
green circle, participants were asked to speak about the subject for 50 seconds, until the circle
was replaced by a red cross. In the baseline trials, which lasted 15 seconds, participants were
asked to recite the nursery rhyme ‘Humpty Dumpty’. This baseline task was designed to invoke
grammatically correct continuous speech without the need for generation of novel utterances.
Each participant responded to 20 prompts.
SL19 Dataset. In this study, participants were asked to produce and comprehend passages of
naturalistic speech while being scanned. They were presented with two runs of the production
and comprehension conditions in an alternating sequence, and the order of the runs was
counterbalanced across participants. Each run lasted approximately eight minutes and
consisted of six speech trials and five baseline trials, randomly presented to each participant.
The duration of the baseline trials was 10s, wherein participants recited or listened to the
Humpty Dumpty nursery rhyme. For the purposes of this analysis, we focused on the speech
production condition. First, in the preparation phase, participants saw a written prompt on
screen for six seconds. Then, they were asked to start speaking at the onset of a green circle,
and to stop when this was replaced with a red cross. Each trial lasted 50 seconds and
participants were asked to try to stay on topic as far as they could. Prior to the experimental
trials, participants were given practice trials to ensure they understood the instructions. Each
participant responded to 12 prompts.

5.2.3. Measures
Spoken responses to each prompt were digitally recorded, transcribed, and cleaned. We
computed measures of coherence, divergence (DSI) and the average number of words per
response.
Global coherence. To quantify the degree to which utterances were related to the topic, we
used a computational linguistic measure of global coherence (Hoffman et al., 2018). This
measure is based on Latent Semantic Analysis (LSA; Landauer & Dumais, 1997), a technique
that uses patterns of word co-occurrence to build a high-dimensional semantic space. These co-
occurrence matrices, generated using large corpora of natural language, track the frequency of
words in different contexts. Singular value decomposition is used to reduce the matrix

127
dimensionality until each word is represented as a high-dimensional vector, with words used
in similar contexts allotted similar vectors. First, for each speech sample, we used LSA to
generate a vector-based representation of the semantic content. Next, we computed a composite
vector by averaging the LSA representations of all participants’ responses to the target prompt
(excluding the target response itself). This composite vector was taken to represent a
prototypical response to the prompt. Finally, for the target response, we constructed smaller
windows of 20 words each, containing the current word and 19 preceding words. For every 20-
word window, we computed an LSA vector to represent the semantic content. Global coherence
was assessed by measuring the cosine similarity of the vector for each window with a vector
representing a typical response to the prompt. Higher global coherence values represented more
typical discourse on the topic, while lower coherence values were suggestive of more off-topic
speech. Global coherence for speech calculated in this way has been shown to be related to
human judgements of coherence (Hoffman et al., 2018). Global coherence values should
theoretically range from 0 to 1, but the range observed in previous studies has been from .20 to
.80.
Divergent semantic integration. To assess the degree to which the responses incorporate
divergent ideas, we used DSI, an established measured of semantic diversity (Johnson et al.,
2022). DSI indexes the semantic distance between ideas in a given response, using a large
language model - bidirectional encoder representations from transformers (BERT) - to generate
context-dependent word vectors. First, BERT processes each response by first splitting it into
sentences and retaining all words and punctuation to provide additional context. For each word
in the response, BERT generates multiple layers of embeddings, using unique sets of weights
to represent the word’s importance in relation to its surrounding context. For example, consider
the word ‘bank’ in the sentences ‘She sat by the river bank’ and ‘She sat with her cheque at the
bank’. BERT generates embeddings for the word ‘bank’ that reflect its’ different meanings
based on the context (‘river’ vs ‘cheque’). These embeddings are then used to calculate DSI:
the embeddings for each word in a sentence are compared to each other using pairwise cosine
semantic distance, and these are averaged across all word pairs in the response. This aggregate
distance provides a single DSI score for the response, representing the degree to which the
response integrates more diverse semantic information. High DSI scores suggest that the
response integrates more divergent ideas, while low scores suggest less semantic diversity. DSI
has been shown to be a good predictor of creativity in narrative text, showing strong
correlations with human ratings of creativity (Johnson et al., 2022; Orwig et al., 2021). DSI

128
values should theoretically range from 0 to 1, but the range observed in previous studies has
been from .70 to .90 (Orwig, Edenbaum, et al., 2024).
No of words. We computed the number of words in each response, as length of the response is
potentially correlated with both DSI (Johnson et al., 2022) and global coherence.

5.2.4. Analysis Strategy

First, we used Pearson’s correlations to understand the relationship between global
coherence and DSI, and their relationship with our covariates of interest, i.e., the number of
words produced and age. To further explore this relationship, we ran an exploratory principal
components analysis on our measures of global coherence, DSI and number of words produced.
An optimal factor solution was derived based on the number of factors suggested by the
Minimum Average Partial method (Velicer, 1976) and Parallel Analysis (Horn, 1965). Finally,
we used a linear mixed effects model to examine whether coherence could predict DSI, after
controlling for age and the number of words in the response. We fit the model with maximum
likelihood estimation and the bobyqa optimiser, using the lme4 package (Bates et al., 2015).
We included random intercepts for participants nested within the datasets, to account for
variation across participants in the different datasets, and random by-prompt intercepts, to
account for variation across the different prompts. Sum to zero coding was adopted for age
group, with younger adults as the first level.
We ran a further exploratory analysis to understand the effect of Task Instruction on
coherence and DSI. We used the MD dataset, as the participants in these studies were either
not given any instruction prior to the trial (‘Baseline’ condition) or were given specific
instructions regarding whether to be focused on the topic (‘Clear’ condition), entertaining
(‘Interest’ condition). We conducted ANOVAs to determine if there were significant differences
in DSI and global coherence across the different task conditions. Post-hoc analyses using
Tukey's HSD test were used to compare specific pairs of conditions. Finally, we ran a linear
mixed effects model to predict DSI scores, using global coherence, age, number of words and
condition as predictors. As above, the model was fit with maximum likelihood estimation and
included random intercepts by-participant and by-prompt. Dummy coding was used for the
task conditions, with Baseline as the reference group.

129
5.3. Study 1: Results

5.3.1. Descriptive and correlational analysis of speech metrics

Over the four datasets, participants generated 2057 responses to 28 distinct prompts
(for a full list of prompts, see Appendix C, Table S8). We excluded nine responses that were
less than 50 words long, leaving us with a final sample of 2048 responses. Descriptive statistics
for the measures of interest are presented in Table 2. Overall, participants produced an average
of 133.43 words per response. Visualisations of the data (see Figure 1c and 1d) indicated that
age may be a potential confound, in line with previous work that suggests older people tended
to be less coherent and produce speech that contained more semantically diverse topics (e.g.,
Hoffman et al., 2018; Wright et al., 2014). Independent sample t-tests supported this
observation, indicating that, compared to their older counterparts, young people produced
speech that was less divergent (t(150.99) = -3.17, p <.01) and more coherent (t(142.01) = 3.58,
p <.001).

Figure 1: Boxplots of DSI and Coherence Across Datasets and Ages. (a) Boxplot of DSI
scores across the five datasets; (b) Boxplot of coherence values across the five datasets; (c)
Boxplot of DSI scores in young and older people; (d) Boxplot of Coherence scores in young
and older people. * = p<.05; ** = p<.01; *** = p<.001

130
 Table 2: Descriptive Statistics for all measures.
Measure Group Mean SD Min Max
MD 0.43 0.08 0.13 0.63
Coherence 18-EL 0.44 0.09 0.20 0.69
19-NC 0.45 0.11 0.19 0.78
19-SL 0.47 0.10 0.23 0.74

Overall 0.44 0.09 0.13 0.78

MD 0.812 0.011 0.779 0.834
DSI 18-EL 0.807 0.011 0.759 0.835
19-NC 0.807 0.010 0.774 0.832
19-SL 0.808 0.009 0.766 0.829

Overall 0.81 0.01 0.76 0.84

MD 146.28 36.83 53 281

No of Words 18-EL 138.03 34.97 50 244
19-NC 103.18 18.50 50 152
19-SL 124.32 25.14 67 196

Overall 133.43 35.38 50 281

Note: SD = Standard deviation, Min = Minimum, Max = Maximum

To examine the associations between the variables, we computed Pearson’s correlations

between the measures of interest in each individual dataset and in the combined dataset (see
Table 3). Overall, coherence was negatively correlated with DSI (r = -.52, p<.001), indicating
a possible trade-off between coherence and divergence. The number of words generated in the
response was negatively related to coherence (r = -.28, p <.001) but positively correlated with
DSI (r = .21, p<.001). Age showed a negative correlation with the number of words generated
(r = -.27, p<.001) and with global coherence (r = -.32, p<.001), but was positively associated
with DSI (r = .27, p<.001). While the negative association between GC and DSI persisted in
the individual datasets, this association did not consistently reach significance.

131
Table 3: Correlation between measures in each individual sample and in the combined
dataset.
Gc - DSI Gc – Words DSI – Words Gc - Age DSI - Age Words-Age

Group r p r p r p r p r p r p
MD -.34 .01 -.28 .04 .46 <.001 -.03 .80 -.01 .94 -.22 .10
18-EL -.68 <.001 -.17 .18 -.12 .36 -.54 <.001 .67 <.001 -.28 .02
19-NC -.32 .23 .34 .20 -.03 .91
19-SL -.26 .19 -.27 .18 .29 .16

Overall -.52 <.001 -.28 <.001 .21 <.001 -.32 <.001 .27 <.001 -.27 <.001

Note: Correlations with age have only been reported in datasets with both young and older adults. Gc = Global
coherence, DSI = Divergent Semantic Integration, Words = Number of words.

Next, we ran an exploratory principal components analysis to understand the underlying

pattern of relationships between the variables. A one factor solution was obtained, accounting
for 56.5% of the total variance in the data (see Table 4). Global coherence had a strong negative
loading, while DSI and number of words loaded positively on this factor. This suggests that
responses that had more semantic diversity and longer lengths were less coherent globally.

Table 4: Principal Components Analysis of Metrics.

Factor 1
Global coherence -.84
Divergent Semantic Integration .81
Number of Words .58

SS Loadings 1.69
Proportion of Variance Explained .57

5.3.2. Global Coherence as a Predictor of Semantic Divergence in Speech

We used a mixed effects model to test whether global coherence predicted DSI after
accounting for confounds, i.e., number of words and age group. The results of the analysis are
presented in Table 5. All variables were scaled prior to inclusion in the model. Coherence had
a significant negative effect on DSI, indicating that increases in global coherence were
associated with a decrease in DSI. Moreover, age was a significant predictor, with younger
individuals demonstrating lower DSI scores overall when compared to the overall mean effect
across age groups (B = -0.33, 95% CI [-0.44, -0.21], p<.001). The number of words generated
was not a significant predictor of DSI. However, the pattern of effects was present in each age

132
group, suggesting that the negative relationship between DSI and coherence holds across
different ages, although younger individuals overall tend to have lower DSI scores.

Table 5: Effects of Coherence, Word Count, and Age on DSI.

Effect

B CI p
Coherence -.13 -.18 - -.09 <.001
Number of Words -.01 -.07 - .04 .67
Age [Young] -.33 -.44 - -.21 <.001

5.3.3. Exploratory Analyses of the impact of Task Instructions on Speech Metrics

We ran an exploratory analysis to examine whether the relationship between coherence
and DSI is influenced by task instructions. In the MD dataset, participants were given specific
conversational goals – they were either told to be entertaining (Interest Condition), clear and
focused (Clear Condition) or were given no specific instructions (Baseline Condition).

Figure 2: DSI and Coherence Across Instruction Conditions. (a) Boxplot of DSI scores
across the three conditions; (b) Boxplot of coherence values across the three conditions

A two-sample t-test revealed no significant differences in DSI between younger adults

(M = 0.81) and older adults (M = 0.81), t(610.7) = −0.484, p=0.63. Next, we performed an
ANOVA to test for differences in DSI across the three instructional conditions (Interest, Clear,
and Baseline). The ANOVA results indicated significant differences in DSI across the different
instructional conditions, F(2,623) = 16.55, p <.001. Follow up post-hoc analyses demonstrated

133
significant differences between the Clear and Baseline conditions (mean difference = -0.0032,
95% CI [-0.0057, -0.0007], p <.001), suggesting that being instructed to be clear results in a
significantly lower DSI compared to the Baseline condition with no specific instructions. The
Interest condition also showed a significant difference from the Baseline condition (mean
difference = 0.0029, 95% CI [0.0004, 0.0054], p=0.02). Participants instructed to be
entertaining were more divergent, compared to the Baseline condition. The most substantial
difference was observed between the Interest and Clear conditions (mean difference = 0.0061,
95% CI [0.0036, 0.0085], p<.001), indicating that participants’ responses in the Interest
condition had a significantly higher DSI compared to those in the Clear condition. These
findings highlight the significant impact of task instructions on DSI, with distinct differences
observed between the Interest, Clear, and Baseline conditions.
We conducted an ANOVA to examine the effect of instructional condition on global
coherence. Results indicated a significant effect of condition on coherence,
F(2,623)=4.59,p=.01. Post-hoc comparisons using Tukey's HSD test revealed that only the
Interest condition significantly differed from the Baseline condition (mean difference = -
0.0236, 95% CI [-0.0423, -0.0049], p = .01), indicating lower global coherence in the Interest
condition. However, the Clear condition did not significantly differ from the Baseline condition
(mean difference = -0.0077, 95% CI [-0.0264, 0.0110], p = 0.6) or the Interest condition (mean
difference = -0.0159, 95% CI [-0.0345, 0.0027], p = 0.11). These results suggest that the
instructional condition has a significant impact on coherence, particularly when comparing the
Interest and Baseline conditions.

Table 6: Effects of Coherence, Number of Words and Condition on DSI.

Predictors
B CI p
Coherence -0.18 -0.26 – -0.10 <.001
Number of Words 0.19 0.10 – 0.28 <.001
Condition [Clear] -0.24 -0.39 – -0.09 0.002
Condition [Interest] 0.16 0.01 – 0.31 0.042

Finally, we ran a linear mixed effects model to examine whether coherence influenced
DSI, after accounting for number of words and the experimental condition. A model with age
group did not improve model fit (F(1) = .34, p=.55), so age was not included in the final model.
Table 6 shows effects of coherence, number of words and condition on DSI. Coherence had a

134
significant negative effect on DSI (B = -0.18, p<.001), indicating that higher levels of
coherence resulted in decreased DSI values. This effect was observed over and above the
influence of the experimental conditions, meaning that global coherence independently
predicts DSI beyond the effects of the task instructions. Conversely, the number of words
generated had a positive effect on DSI (B = 0.19, p<.001). When compared to the baseline
condition (no specific instructions), the clear condition negatively predicted DSI (B = -0.24, p
= 0.002), suggesting that participants instructed to be clear tended to produce less semantically
divergent information. In contrast, the interest condition positively predicted DSI (B = 0.16, p
= 0.042), indicating that when participants are instructed to be entertaining, they tend to
generate more diverse speech.

5.4. Study 1: Discussion

In Study 1, we first examined whether global coherence predicts the Divergent
Semantic Index (DSI). Our analysis of 2054 responses across four datasets revealed a
significant negative association between coherence and DSI, suggesting that coherence and
divergence in narrative speech are inherently linked, with higher coherence generally leading
to lower divergence and vice versa. This trade-off highlights the complex balance individuals
must maintain between generating divergent ideas and producing structured, on-topic speech.
It also highlights the importance of controlling for divergence when assessing effects of
coherence and vice-versa. The significant impact of task instructions on both DSI and GC
further indicates that external goals can modulate this balance. Specifically, instructions to be
clear lead to less divergent speech, while instructions to be entertaining result in more divergent
and less coherent speech. However, the effect of coherence on DSI persisted regardless of task
condition, indicating that coherence independently predicts divergence. Building on these
insights, Study 2 aims to investigate the neural mechanisms underpinning the observed trade-
off between coherence and divergence. Using functional connectivity analyses, we will
investigate how different brain networks interact to support creative and coherent speech.
Specifically, we will examine the roles of the Default Mode Network (DMN), the Multiple
Demand Network (MDN) and the Semantic Control Network (SCN) in mediating the balance
between divergent and coherent thought processes.

5.5. Study 2: Methods

Study 2 reports analyses of the 19-SL dataset, originally collected for my MSc
dissertation. Other analyses of these data have been reported previously (see Morales et al.,
2022; Patel et al., 2023), but this is the first study to examine the effects of DSI.

135
5.5.1. Participants
25 participants (MAge= 24, SDAge = 4.4, range = 18 -35, 21F, 4M) were recruited through
the University of Edinburgh and compensated for their time. Participants were right-handed
(based on the Edinburgh Handedness Inventory; Oldfield, 1971); native English speakers
(defined as learning English before age 5); and in good health, with no history of neurological
or psychiatric illness. Ethical approval was received from the Edinburgh Psychology Research
Ethics Committee and all participants gave informed consent.

5.5.2. Design and Procedure

Participants were required to either generate or listen to naturalistic speech in 50 second
blocks while in the scanner. The experiment consisted of four runs in total, with each participant
completing two runs of production and comprehension respectively. The order of the runs was
counterbalanced across participants, and each run lasted approximately eight and a half
minutes. For every run, there were six discourse trials and five baseline trials, randomised for
each participant. The current study analysed the speech production runs: for each trial,
participants were presented with the prompt on a screen for six seconds, to allow them to
prepare their speech. They were instructed to start speaking when a green circle appeared in
the centre of the screen, and to stop when a red cross appeared at the end of the 50 second
period. In the baseline condition, participants recited the nursery rhyme for 10 seconds and
were asked to start again from the beginning if they finished early (see Figure 3a). The
interstimulus interval between trials was jittered between 3-7 seconds (Mean = 5 seconds). The
stimuli were presented using E-prime software on a screen behind the scanner, viewed on a
mirror placed on the head coil. Participants were given practice trials at the start to familiarise
them with the production task and the baseline condition.

5.5.3. Materials and Measures

To elicit speech production, we devised 12 prompts related to common semantic
knowledge on a series of topics (e.g. Do you think the internet has improved people’s lives?
See Appendix C, Table S8 for a list of all prompts used). These semantic speech conditions
were contrasted with a baseline condition, in which participants recited the popular nursery
rhyme ‘Humpty Dumpty’. This active low-level baseline was chosen as it consists of
grammatically sound yet automatic speech that required minimal semantic processing. In
addition to the two speech production runs, there were two additional runs of speech
comprehension that have not been analysed in the present study.

136
Figure 3: Workflow from Data Collection to Neural Analysis. (a) Illustration of a
production/baseline trial; (b) Transcription; (c) Computation of Coherence values; (d)
Computation of DSI; (e) Constructing model predictors; (f) Stages in Neural Analysis. Note:
Image (c) has been adapted from Hoffman (2019).

5.5.4. Processing of speech samples

Figure 3 illustrates the stages in the analysis pipeline. Speech was recorded using an
MRI-compatible microphone, processed with noise cancellation software to reduce scanner
noise (Cusack et al., 2005) and transcribed for analysis.

137
Global coherence. To index global coherence, we used the same computational linguistic
measure outlined in the Methods of Study 1. Responses to each prompt were divided into five-
second blocks, transcribed, and cleaned of contractions and punctuation. For every response to
a target prompt, we computed a vector-based representation of the semantic information using
Latent Semantic Analysis (LSA; Landauer & Dumais, 1997). These representations, excluding
the target response, were aggregated to form a composite vector reflecting a prototypical
response to that prompt. Using a moving-window approach, we compared the LSA vectors for
every 20-word chunk in a response to this composite vector. The cosine similarity between
these vectors was taken to indicate the degree of semantic similarity between the content in
each window and the prototypical response. Values ranged from 0 to 1, with higher values
implying that the speech was very strongly related to the prompt. To derive a single global
coherence value, we averaged these cosine similarities across all windows in the speech
sample. This global coherence score was then used as a predictor of BOLD responses and
connectivity (Hoffman, 2019).
Divergent semantic integration. As in Study 1, we used DSI to index the degree to which each
response connected divergent ideas (Johnson et al., 2022). DSI uses bidirectional encoder
representations from transformers (BERT) to generate context-dependent word vectors to
measure the semantic distance between ideas in a response. To extract these vectors, the process
was as follows: first, BERT split each response into sentences, and generated layers of
embeddings for each word in the response. Each embedding assigned different weights
depending on the importance of the word in relation to the surrounding words, capturing the
nuances of how meaning changes depending on context. Next, pairwise cosine semantic
distance between word embeddings was computed, and this was averaged across all words in
the response. The average distance was taken to reflect a single DSI score for the response,
indicating the degree to which the response incorporated diverse semantic information.
No of words. We computed the number of words in each response, as length of the response
may be a potential covariate affecting DSI scores (Johnson et al., 2022) and global coherence.

5.5.5. Image Acquisition and Processing

Participants were scanned in a 3T Siemens Prisma Scanner using a 32-channel head coil.
Functional image data were acquired at three echo times (13ms, 31ms and 48ms) using a whole
brain multiband multi-echo acquisition protocol (Feinberg et al., 2010; Moeller et al., 2010; Xu
et al., 2013). This multi-echo protocol is useful when dealing with speech-related head
movements and potential signal drop-out in the ventral temporal regions. Pre-processing was

138
carried out using Statistical Parametric Mapping 12 (SPM 12, v7219) and TE-Dependent
Analysis Toolbox 0.0.7 (tedana; DuPre et al., 2021). The three-echo series were weighted and
combined into a single time series, which was divided into BOLD components and noise using
Tedana. This is an approach that has been used to improve signal quality and reduce
susceptibility artifacts commonly seen in the Anterior Temporal Lobes (ATL; Kundu et al.,
2017). 46 image slices covering the whole brain were obtained with a TR = 1.7s, 80 x 80 matrix,
flip angle = 73°, and isotropic voxel size of 3mm. 281 volumes were acquired over each run.
We also acquired T1-weighted structural images using an MP-RAGE sequence with 1mm
isotropic voxels, TR = 2.5 s, TE = 4.6 ms, flip angle = 7°, in-plane resolution = 224 × 304
matrix, 0.8mm x 0.8mm. 288 slices of slice thickness 0.8mm were acquired (no slice gap). Pre-
processed images were then unwarped and spatially normalised to Montreal Neurological
Institute (MNI) space using SPM DARTEL (Ashburner, 2007), which allows for precise inter-
subject registration and increased analytic sensitivity. These spatially normalised images were
then smoothed using Gaussian kernels with 8mm full width at half maximum.

5.5.6. Analyses
Subject-level models. First, we estimated subject-level models to get individual level beta maps
representing the effects of discourse characteristics on activation. Four event types were
modelled for each run: semantic prompts and baseline prompts (6 second duration); semantic
discourse (50 second duration) and baseline speech (10 second duration). For the semantic
discourse events, three parametric modulators were used, representing the global coherence,
DSI and number of words for each response. Additionally, 12 motion parameters were added
to the model as covariates. These were convolved with the haemodynamic response function
using SPM12. To assess the brain regions involved in semantic speech, we contrasted brain
activation for semantic speech against the baseline task. To assess effects of discourse
characteristics on activation, we used the three parametric modulators of the discourse effect.
Data were treated with a high-pass filter with a cut-off of 128s.

Whole-brain level. First, we investigated whole brain activation at the group-level. For the
semantic vs baseline speech contrast, we used a voxel-level threshold of p<.001 with correction
for multiple comparisons at the cluster level using random field theory. For the effects of global
coherence, DSI, and number of words, few regions were significant at this stringent threshold.
Therefore, to increase power, our main analyses of these effects focused on three large-scale
brain networks, described in the following sections. However, to visualise these effects across
the whole brain, we present unthresholded beta maps in Figure 6, to visualise the pattern of the

139
effects. Note, the main conclusions of these effects are drawn from the network activation and
functional connectivity analyses described next.

Networks of Interest. Our analyses were performed on three networks of interest that have
been shown to be related to both coherence and creativity. The networks and masks used to
identify them were the same as those used in (Morales et al., 2022):

(1) The Default Mode Network (DMN), identified from a 7 network parcellation of resting-
state fMRI data from 1000 participants reported by Yeo et al., (2011). The DMN is
shown in blue in Figure 4, and includes regions of the medial prefrontal cortex,
posterior cingulate cortex, anterior temporal lobe, precuneus, and lateral parietal cortex.
(2) The Semantic Control Network (SCN), defined using significant voxels identified in a
meta-analysis by Jackson (2021), which highlights regions consistently associated with
high cognitive control demands in semantic processing tasks. SCN regions are shown
in green in Figure 4, and include the left inferior frontal gyrus, the left posterior middle
temporal gyrus, and presupplementary motor regions.
(3) Multiple-Demand Network (MDN), defined as the regions associated with increased
cognitive demands across multiple domains, as reported by (Fedorenko et al., 2013).
The MDN is shown in red in Figure 4, and includes regions in the dorsolateral PFC,
precentral gyrus, insular cortex, intraparietal sulcus, presupplementary and
supplementary motor areas, anterior and middle cingulate cortex.

Figure 4: Masks of the networks of interest. Note: Blue = DMN; Green = SCN; Red = MDN

Identification of networks using Independent Components Analysis (ICA). The main aim of
this analysis was to understand the effects of discourse coherence, DSI and number of words
on activation in our three networks of interest. To identify and extract the BOLD timecourses
in our networks of interest, we used Independent Components Analysis (ICA), a data reduction
technique that separates fMRI signals into spatially or temporally independent components that

140
correspond to different brain networks (Calhoun et al., 2009). We used a spatially-constrained
semi-blind version of ICA, which extracts independent components from the data that align
with our pre-specified spatial templates (Lin et al., 2010). This constrained approach ensures
prior spatial information about the expected networks is incorporated into the estimation
process, and that we do not have to make post-hoc judgements about how many or which
components to analyse. This was implemented in Matlab using the Group ICA of fMRI toolbox
(GIFT; [Link] We ran the spatially-constrained ICA separately
on each participant’s data, using our pre-determined network masks to constrain the extraction
of components. This was done to ensure we recovered exactly three components from each
participant’s data, corresponding to our networks of interest.

Activation Analyses. Next, we ran a series of general linear models to investigate the effects of
coherence and DSI on activation in each network using the BOLD timecourses extracted from
the GIFT ICA analysis. Data were treated with a high pass filter with a cut-off of 180s. The
regressors in the GLM were identical to those used in the whole-brain analyses, but here, they
were used to predict the BOLD response in each network, rather than the response in individual
voxels. Contrast values representing the difference in activation between semantic and baseline
speech and the effects of GC, DSI and words were computed for each network. Data were
aggregated by averaging over the two production runs. FDR-corrected t-tests were conducted
on the effects in each network to determine whether these were significantly different from
zero. We tested whether effects differed between networks using one-way repeated measures
ANOVAs, with network type (DMN, MDN, SCN) as the within-subjects factor, and
participants as the random effect. Sum to zero coding was used for network type. Following
this, pairwise comparisons using paired t-tests with Holm adjustment were used to compare
networks.

Functional connectivity analyses. Next, we wanted to explore how the functional connectivity
between our networks of interest were influenced by the coherence and creativity of the
responses participants made. Psychophysiological Interactions (PPI) is a popular type of
functional connectivity analysis, which evaluates changes in connectivity strength induced by
specific cognitive tasks or conditions. We used the generalised psychophysiological
interactions method (gPPI; McLaren, Ries, Xu, & Johnson, 2012) that allows for multiple PPI
analyses within a single general linear model. The gPPI method is a powerful tool that identifies
brain regions showing task-dependent functional connectivity with a seed region of interest.
Specifically, it examines how the functional coupling between a seed region and other brain

141
regions change as a function of experimental condition(s). We utilized this method to
understand how the strength of the correlations between our three large-scale networks of
interest varied as a function of our cognitive predictors.
While traditional PPI analyses use small, predefined seed regions, we were interested
in functional interactions between large-scale networks. For our seeds, we used the BOLD
timecourse data extracted for each network of interest. This approach is similarly adopted in
(Goulden et al., 2014; Masson et al., 2020). We used the gPPI toolbox to generate task, seed,
and PPI regressors for our model using each network as a seed. We then used these models to
predict the timecourses in a target network, e.g., DMN-seeded models were used to predict the
timecourses of the SCN and MDN. These analyses demonstrated how the between-network
correlations were influenced by experimental tasks. PPI effects are considered to be bi-
directional, so effects from network pairs (e.g. DMN-to-SCN and SCN-to-DMN) were
averaged before conducting our analyses.
We tested the effects of our predictors on the functional connectivity between our ICA-
derived networks. First, t-tests (FDR-corrected) were used to determine whether the functional
connectivity effects between each network pair were significantly different from zero. Next,
we ran repeated measures one-way ANOVAs, with network pairs as the sum-to-zero coded,
within-subject variable and participants as the random effect. Pairwise comparisons using
paired t-tests with Holm adjustment were then conducted to compare network pairs.

5.6. Study 2: Results

The 19-SL dataset originally contained 300 responses, but 3 responses were removed
as they contained less than 50 words, resulting in a final sample of 297 responses. The
average response length of 124.32 words (SD = 25.14), with a mean global coherence score
of 0.47 (SD = .10), and a mean DSI score of 0.808 (SD = .009).

5.6.1. Whole Brain Analyses

We ran a whole brain analysis to examine the regions that were activated in relation to
our regressors. First, we examined brain activation for semantic versus baseline speech (Figure
5). The production of semantic speech activated areas typically associated with semantic
processing, including the bilateral temporal poles, superior and middle temporal gyri, superior
and inferior frontal gyrus (BA 45 and 47), and the angular gyrus (Binder & Desai, 2011). There
was also activation in the medial occipital area, which may have been related to reading the
prompts, and the cerebellum, which has been associative with predictive linguistic processing.

142
Figure 5: Whole brain activation for semantic compared to baseline speech. Note:
Images thresholded at cluster-corrected p<.05.

Figure 6: Unthresholded maps (beta values) of the effects of the regressors on neural
activation. (a) Coherence effects; (b) DSI effects; (c) No of words effects

143
 For the effects of global coherence, DSI and number of words on activation, few regions
were significant at this stringent threshold. Therefore, to increase power, our main analyses of
these effects focused on three large scale brain networks, described in the following sections.
However, to visualise these effects across the whole brain, in Figure 6, we present the
unthresholded beta maps that show the direction and magnitude of the effects of the regressors
in each region. Note, these maps are shown to only visualise the patterns of the effects, and our
main conclusions are drawn from the network analyses described next.

For coherence effects, regions shown in red (Figure 6a) showed positive activation
when participants generated more coherent speech. These regions included the classic
executive control regions, such as the anterior cingulate cortex, supramarginal gyrus, precentral
gyrus, in addition to the inferior frontal gyrus and the insula, and the middle frontal gyrus.
Conversely, regions in blue showed negative effects, showing more activation to less coherent
speech. These regions included parts of the DMN such as the temporal poles and regions of the
superior frontal gyrus. In contrast, the temporal regions, particularly the inferior and anterior
temporal lobes, showed positive activation for more divergent speech (Figure 6b), while
regions related to cognitive control showed activation to less divergent speech, including parts
of the frontal gyrus. Finally, a distributed set of regions showed positive activation in relation
to the number of words produced in the response, including the angular gyrus, postcentral
gyrus, precuneus, middle and superior frontal gyrus and the middle temporal gyrus.

144
5.6.2. Network Activation Analyses
Table 7: One-sample t-tests for effects across each ICA-derived network.
Dependent Variable Network t p Adjusted p
Semantic vs Baseline DMN 5.08 <.001 <.001
MDN -6.53 <.001 <.001
SCN 8.66 <.001 <.001

Coherence DMN -2.12 .05 .06

MDN 3.85 <.001 <.001
SCN 1.95 .06 .06

DSI DMN -1.09 .29 .43

MDN -.70 .49 .49
SCN -1.78 .09 .27

No of Words DMN -.40 .70 .96

MDN -.05 .96 .96
SCN -.49 .63 .96

The DMN and SCN showed increased activation during semantic speech, while the
MDN showed more activation to the baseline speech condition. As shown in Table 7, all FDR-
corrected one sample t-tests indicated that three contrasts were significantly different from
zero: the DMN (t (46)=5.08, p<.001) and SCN (t (46)= 8.66, p<.001) contrasts were
significantly greater than zero, indicating increased activation during semantic speech
compared to baseline; in contrast, the MDN contrast (t (46)=-6.53, p<.001) was significantly
less than zero, indicating decreased activation during semantic speech compared to baseline.
These findings suggest a distinct pattern of neural activation in response to semantic speech
compared to baseline, in line with previous findings (Morales et al., 2022; Patel et al., 2023).
The ANOVA results demonstrated a significant main effect of network, F(2, 46) = 88.87,
p<.001. As shown in Table 8, all pairwise comparisons were statistically significant, indicating
that each network had significantly different contrast values to the other.

145
Table 8: Paired t-tests for Network Differences Across Semantic and Coherence Measures

Dependent Variable Network Comparison Adjusted p

Semantic vs Baseline DMN - MDN <.001

DMN – SCN <.001

MDN – SCN <.001

Coherence DMN - MDN <.001

DMN – SCN <.001

MDN – SCN .34

DSI DMN - MDN 1

DMN – SCN 1

MDN – SCN 1

No of Words DMN - MDN 1

DMN – SCN 1
MDN – SCN 1

For global coherence, FDR-corrected t-tests indicated that activation in the MDN for
coherent speech was significantly different from zero, indicating greater engagement for the
MDN when people produce more coherent speech. Effects in the SCN and DMN did not
survive correction for multiple comparisons. Further, the repeated measures one-way ANOVA
revealed a significant effect of network type, F (2,46)=17.32, p<.001. Pairwise revealed
significant differences between the DMN and both the MDN and SCN, but not between the
MDN and SCN (adjusted p = .34). The significant difference between the DMN, compared to
the MDN and SCN, suggests this network plays distinct role in relation to maintaining speech
coherence. The MDN and SCN did not differ significantly, indicating there may be some shared
functional similarities in their roles. In line with this view, the DMN showed decreased
activation while people were being more coherent, while the MDN and SCN showed increased
activation when people were being coherent.
With regard to DSI, t-tests revealed that activation did not significantly differ from zero
in any of our networks of interest. Further, the effects of the repeated measures one-way

146
ANOVA did not find an effect of network on DSI, F(2,46) = .27, p=.72), or the number of
words, F(2,46) = .06, p=.94. This indicates there was no significant differences in effects across
our networks of interest in relation to DSI and words in the response.

Figure 7: Effects of regressors in each network. (a) Semantic vs Baseline speech; (b) Global
Coherence; (c) DSI; (d) Number of Words. *** = p <.001

5.6.3. Functional Connectivity Analyses

For the semantic vs. baseline contrast, one sample t-tests (FDR corrected) indicated that
that the PPI contrasts for each network pair were not significantly different from zero. These
results are presented in Table 9. The ANOVA further confirmed there were no significant task-
dependent changes in the functional connectivity between our networks of interest, F(2, 46) =
1.06, p=.35. Taken together, these results that the experimental task did not alter the functional
coupling between networks in our study (Figure 8a).

147
Table 9: One sample t-tests for effects in each network pair.
Dependent Variable Network t p Adjusted p
Semantic vs Baseline DMN-MDN .26 .80 .80
DMN-SCN -.93 .36 .54
MDN-SCN -1.90 .07 .21

Coherence DMN-MDN .03 .98 .98

DMN-SCN -1.05 .30 .46
MDN-SCN 1.12 .28 .46

DSI DMN-MDN 3.53 <.01 <.01

DMN-SCN -.02 .98 .99
MDN-SCN -.01 .99 .99

No of Words DMN-MDN 1.64 .12 .17

DMN-SCN 2.35 .03 .08
MDN-SCN .41 .68 .68

Next, we tested whether there is a difference in functional connectivity between brain

networks during more versus less coherent speech. T-tests indicated PPI contrasts were not
significantly different from zero after the FDR-correction (as seen in Table 9). The ANOVA
revealed no significant differences in connectivity between network pairs, F(2,46) = 2, p =.14.
This implies that speech coherence is not correlated with changes in functional connectivity
between networks in our study (Figure 8b).
Next, we tested whether there is a difference in functional connectivity between brain
networks when people were being more divergent. The FDR-corrected t-test indicated that the
DMN-MDN effect was significantly greater than zero t(2, 46) = 3.53, p <.01, suggesting that
being more divergent increases the connectivity between the DMN and MDN (Figure 8c).
Further, the ANOVA suggested that there were differences in functional connectivity between
network pairs, F(2,46) = 3.73, p=.03. However, this effect did not appear in the pairwise
comparisons.

148
 Table 10: Paired t-tests for Network Differences in Semantic, Coherence, DSI, and Word
Count Measures.

Dependent Variable Adjusted

Network Comparison
p
Semantic vs Baseline DMN-MDN – DMN-SCN .79
DMN-MDN – MDN-SCN .16
DMN-SCN – MDN-SCN .79

Coherence DMN-MDN – DMN-SCN .65

DMN-MDN – MDN-SCN .65
DMN-SCN – MDN-SCN .09

DSI DMN-MDN – DMN-SCN .11

DMN-MDN – MDN-SCN .11
DMN-SCN – MDN-SCN .99

No of Words DMN-MDN – DMN-SCN .90

DMN-MDN – MDN-SCN .90
DMN-SCN – MDN-SCN .59

Finally, when people produced more words, none of the network pairs showed
activation that was significantly different from zero. The ANOVA indicated no differences
between network pairs F(2,46) = .85, p=.43.

149
Figure 8: Effects of regressors on network connectivity. (a) Semantic vs Baseline speech; (b)
Global Coherence; (c) DSI; (d) Number of Words. ** = p <.01

5.7. General Discussion

The aim of this work was to explore the relationship between coherence and creativity
in spontaneous speech. Specifically, we wanted to understand how speakers strike a balance
between sustaining a logical flow of ideas and incorporating originality into unscripted verbal
communication. To this end, we conducted two studies examining this relationship at the
behavioural and neural levels. In the first study, we collated a large set of verbal responses to
topic prompts and used established methods to compute measures of global coherence and
creativity (Divergent Semantic Integration) in speech. We found that the two constructs are
negatively related: higher coherence was related to less divergence in speech, suggesting that
there is a trade-off between being coherent and creative. In the second study, we examined the
neural mechanisms underlying these two constructs, using Independent Components Analyses
to isolate three networks that have been linked to both coherence and creativity (i.e., the Default
Mode, Multiple Demand and Semantic Control Networks). Following this, network activation
analyses indicated that more coherent speech was related to increased activation in the MDN.

150
Further, functional connectivity analyses suggested that divergence in speech was underpinned
by an increased coupling between the DMN and MDN. Together, these findings highlight a
complex interplay between these two constructs in spontaneous speech. To our knowledge, this
is the first study to investigate creativity as an emergent, everyday quality within spontaneous
speech, a context that has not been explored in traditional creativity assessments.
First, our behavioural study identified a negative relationship between global coherence
and divergence in extemporaneous speech. This suggests that when speakers prioritise
maintaining a coherent and structured narrative, it may come at the expense of the limiting the
originality and diversity of ideas in their speech. Moreover, we found that both coherence and
creativity can be modulated as per the conversational goal: when participants were asked to be
clear, their speech was less divergent and more coherent; and they displayed the opposite
pattern when prompted to be entertaining. Importantly, however, the negative relationship
between coherence and DSI persisted even after controlling for this instructional condition,
highlighting the effects of coherence on divergence irrespective of the specific task instructions
given. These findings suggest that coherence and creativity may function as two opposing
forces in communication, where enhancing one may naturally diminish the other. This trade-
off between coherence and creativity is echoed in previous work examining creative story
writing: stories that had greater global cohesion, reflecting the degree to which ideas were
connected to the central narrative, were rated as being less original (Fan et al., 2023). Taken
together, we can surmise that the mental resources required to regulate speech coherence – such
as organising thoughts and maintaining a logical sequence of ideas –may limit the speaker’s
ability to introduce unconventional or unrelated ideas. Thus, prioritising coherence might lead
to more predictable speech patterns, whereas allowing for divergence may lead to a more
creative but potentially less organised narrative.
The optimal balance between coherence and divergence in speech may differ based on
the context and purpose of communication. Greater coherence may be helpful in contexts
where one needs to communicate ideas efficiently and accurately, with any divergence leading
to a detraction from the main message. However, if the goal of the conversation is to be
interesting, then a more meandering conversational style that includes irrelevant details, off-
topic comments or unexpected twists could engage the audience more effectively, making the
interaction more stimulating. Indeed, this trade-off impacts how the audience perceives the
speaker, with listeners rating less coherent speakers as being more interesting (James et al.,
1998).

151
 This balancing act between coherence and divergence may be closely tied to the
speaker’s executive control abilities. Our behavioural results indicated that older people tended
to be less coherent and more divergent in their speech, replicating previous findings in the
literature that suggest that older people are more likely to produce off-topic speech (Arbuckle
& Gold, 1993; Glosser & Deser, 1992; Hoffman, Loginova, et al., 2018; Kemper et al., 2010;
Kintz et al., 2016; Marini et al., 2005; Marini & Andreetta, 2016; Wright et al., 2014). This is
thought to stem from age-related declines in domain-general cognitive control mechanisms,
leading to the implication of these processes in maintaining coherence in speech. Aligning with
this view, previous work has emphasized the importance of domain-general executive control
processes for speech coherence, demonstrating that more coherent speech is associated with
superior attentional and control abilities (Arbuckle & Gold, 1993; Hoffman et al., 2018;
Kemper et al., 2010; Kintz et al., 2016; Marini & Andreetta, 2016; Wright et al., 2014b). To
produce meaningful utterances, one must generate relevant and interesting statements while
continuously monitoring the flow of ideas. Here, top-down control is required to structure ideas
in a logical order, ensuring the narrative remains clear; and to maintain attentional focus on the
main topic, facilitating the smooth integration of subtopics and related points.
In line with this proposition, our network activation analyses indicated that greater
speech coherence was related to increased activation in the Multiple Demand Network, a
system that has been implicated in higher-level cognitive processes that organise, manage, and
execute sub-tasks to support the accomplishment of a broader objective (Duncan, 2010;
Fedorenko et al., 2013). This suggests that the role of the MDN in facilitating speech coherence
may lie in keeping the speaker on track by managing sequential operations crucial for the
temporal organisation of speech. These operations include planning the sequence of ideas,
ensuring all parts of speech contribute to the overall narrative, transitioning between ideas
effectively to maintain logical flow, suppressing irrelevant thoughts or distractions that could
lead to off-topic ideas, monitoring speech for errors, and redirecting speech back to the topic
by adjusting content in real time. Indeed, beyond classical language areas, speech production
has been linked to a network of regions associated with cognitive control (AbdulSabur et al.,
2014; Adank, 2012; Awad et al., 2007; Stephens et al., 2010). This aligns with the notion that
these processes are necessary for managing the complex task of generating meaningful,
connected speech (Bourguignon et al., 2018; Bourguignon & Gracco, 2019; Geranmayeh et al.,
2014, 2016; Wise & Geranmayeh, 2016). The importance of domain-general cognitive
mechanisms for speech regulation has also been illustrated in neuropsychological studies (e.g.,
Barker et al., 2017; Marini, 2012; Marini et al., 2011). For example, Barker et al., (2017) found

152
that mild stroke patients with right hemisphere damage exhibited significantly lower global
coherence, and moreover, they exhibited attentional and executive impairments that were
associated with difficulties in producing coherent speech. Specifically, impairments in
selectively attending to relevant information and in inhibiting irrelevant distractors was linked
to greater difficulties in maintaining the overall structure and relevance of their discourse. The
findings of the current study add to this literature, underscoring the critical involvement of the
MDN in orchestrating executive and attentional processes necessary for sustaining speech
coherence, and further highlighting the integration of domain-general cognitive mechanisms
with language-specific functions to facilitate speech production (Bourguignon & Gracco,
2019).
However, it is important to note that the findings in our study differs from that of
Morales and colleagues (2022) who, using the same dataset, did not find that the MDN was
modulated by speech coherence during production or comprehension. The discrepancy in our
results may be attributed to differences in the approach to computing coherence values. Morales
et al. (2022) tracked coherence over time, analysing how it fluctuated throughout the speech
production process. In contrast, our approach focused on computing a single, aggregated global
coherence value to represent the entire response. The differing results suggest that coherence
may be particularly sensitive to the temporal resolution of the analysis. Consequently, future
studies should investigate how varying the temporal windows of assessment influences
coherence, as this could offer deeper insights into the temporal dynamics of speech coherence
and its neural underpinnings.
The present work did not strongly implicate the semantic control network in coherence,
which might suggest a deviation from prior research suggesting semantic control processes
play a pivotal role in maintaining global coherence, beyond domain general control. For
instance, previous neuropsychological work has demonstrated that patients with semantic
aphasia, who exhibit deficits specific to semantic control, struggle to maintain global
coherence, but can manage local coherence (Hoffman et al., 2020). Additionally, Hoffman et
al., (2018) highlighted the association between semantic selection mechanisms, the ability to
focus on specific featural properties while avoiding dominant associations, and global
coherence. In natural discourse, a conversational cue may activate a wide of range of semantic
information, not all of which is relevant to the conversational goal. Enhanced semantic
selection mechanisms may allow the speaker to effectively select those aspects of knowledge
that are pertinent to the conversation at hand. Supporting this, neuroimaging research in healthy
older adults has linked increased coherence to activation in the LIFG - and more specifically,

153
the pars triangularis (BA 45) - a key area of the semantic control network linked to semantic
selection mechanisms (Hoffman, 2019). Notably, the literature points to a more domain-general
role for BA 45 in resolving competition, extending beyond semantics to episodic and working
memory domains (Badre & Wagner, 2005; Dobbins & Wagner, 2005; Jonides & Nee, 2006).
However, if we consider evidence at the network level, the picture becomes slightly
more complex. Specifically, Morales et al. (2022) found that listening to less coherent speech
was associated with increased activation in the LIFG, supporting the idea that semantic control
regions regulate the selection of ideas during naturalistic speech. Notably, though, this
engagement was task dependent, as SCN activity was related to less coherent speech only
during speech comprehension, but not during production. This observation suggests that the
role of the SCN in regulating speech coherence may be more nuanced than previously
recognized, stressing the need for further investigation into when and how this regulatory
process occurs. Specifically, future work could explore whether the SCN is differentially
recruited when speech tasks vary in cognitive demand, highlighting the specific conditions
under which semantic control regions are necessary for coherent speech.
Next, we turn to the question of creativity in spontaneous speech. Our functional
connectivity analyses highlighted a dynamic interplay between the default and executive
networks in the context of more divergent speech patterns. This finding suggests that the
production of more semantically diverse speech is linked to interaction among a broad set of
interconnected networks. Typically, the DMN is associated with internally directed cognition,
such as daydreaming or mental simulation (Andrews-Hanna et al., 2018; Andrews‐Hanna et
al., 2014; Fox et al., 2015); while the MDN is linked to task-focused executive functions
(Duncan, 2010; Fedorenko et al., 2013). Our results indicate that these networks may work in
tandem when people access more semantically diverse speech content. This default-executive
interplay suggests that divergent thinking requires processes to aid the generation of novel
ideas, as well as organisational control to manage and apply these ideas in a meaningful way.
This aligns with numerous studies that have proposed that default and executive networks
exhibit dynamic coupling during divergent thinking tasks, reflecting the complex interaction
between associative and executive processing underlying creativity (e.g., Beaty, Benedek, et
al., 2014; Beaty et al., 2015; Beaty, Silvia, et al., 2014). Building on this, the results of the
current study extend our understanding of the neural underpinnings of creativity, illustrating
how these networks operate beyond controlled experimental tasks into the realm of
spontaneous everyday speech.

154
 However, it is important to consider the nature of the prompts used in our study, as they
may have shaped the balance between coherence and creativity in the speech produced. Across
datasets, we used a variety of prompts to elicit speech on general semantic knowledge. Some
prompts were open-ended (e.g., ‘What would it have been like to live in the Middle Ages?’),
encouraging more creative and expressive speech. In contrast, procedural prompts (e.g., ‘How
would you make a cup of tea or coffee’) may have activated well-established scripts of a
sequence of events. As a result, speech produced in response to these prompts may have been
more coherent, but at the expense of creativity. The 19-SL dataset, which forms the basis of
Study 2, contained many procedural prompts (e.g., "Describe how you would make a cup of
tea or coffee", "Describe a typical visit to a grocery store", "Describe the steps you’d take to
order-in food") that may have been different from the more open-ended, exploratory questions
used in other datasets. As a result, responses in this dataset were likely to be more coherent, as
participants may have relied on familiar scripts from memory. This aligns with the findings
from (Trunk & Abrams, 2009), who observed that participants tended to favour logical
structure and comprehensibility for procedural topics, whereas more episodic topics (‘Favorite
vacation’) encouraged more entertaining/expressive dialogue. These differences suggest that
the type of prompt may systematically shape the balance between coherence and creativity in
speech production, which is important to consider when interpreting our results.

Interestingly, our study did not find that the Semantic Control Network was related to
divergence in speech content. One potential reason for this could be that participants might not
have been sufficiently taxing their semantic systems, potentially relying on automatic retrieval
from episodic memory instead. While the content generated could be considered “semantically
diverse” in comparison to other participants’ responses, the information might not have been
novel to the participant themselves. Indeed, while our DSI measure represented the semantic
distance between ideas within the response, it does not capture whether the content is novel to
the individual themselves. Additionally, given that many of the prompts required participants
to talk about familiar and well-established scenarios, this may have invoked episodic memories
and familiar schemas, reducing the demand on the semantic control processes and thereby
influencing the degree to which the SCN was engaged. For example, when asked ‘What sort of
things usually happen at a wedding?’, participants may effortlessly dip into their mental storage
vaults of information related to past weddings, recounting details from personal experiences
rather than crafting new narratives.

155
 This distinction between automatic and controlled retrieval is crucial, as the SCN is
typically engaged when the retrieval process requires more effort (e.g., Chiou et al., 2018; Gao
et al., 2021; Whitney et al., 2011b; Zhang et al., 2021). In line with this, Krieger-Redwood et
al. (2023) found that the SCN is preferentially recruited when participants are required to
generate unconventional links between items (melon-bookcase), indicating a greater demand
for controlled retrieval processes. In contrast, strong associations (red-round), are supported by
automatic, associative patterns of retrieval underpinned by DMN activity. Critically, as in our
study, these strongly associated items may evoke a range of episodic as well as semantic
information, suggesting a more integrated form of memory retrieval from past experiences as
well as well-established sources of knowledge. For example, when presented with ‘red-round’,
one might make the association ‘clown nose’, drawing on common knowledge and personal
memories of clowns. This integration allows for more automatic recall, in contrast to the
effortful search required to make connections between more distantly related items. To sum up,
the lack of SCN involvement in our study could indicate that the participants’ speech content
was been rooted in familiar or easily accessible knowledge, thereby reducing the need for active
engagement of the SCN’s regulatory functions.
Furthermore, Krieger-Redwood et al. (2023) highlighted a functional dissociation
within the DMN: the ‘core’ DMN system (including regions such as AG, PCC, and ACC) and
the dorsomedial subsystem (dmDMN; consisting of the ventral IFG, temporal and parietal
regions). The core DMN is typically implicated in tasks that rely on strong semantic
associations and episodic memory, such as autobiographical and self-referential thinking. In
contrast, the dorsomedial subsystem is engaged in more effortful, controlled retrieval, such as
when semantic control is required to navigate the weaker links between concepts. This
distinction underscores the flexibility of the DMN in supporting both automatic and controlled
processing, depending on the complexity of the task. Crucially, the dmDMN overlaps with the
SCN, suggesting these networks are functionally interconnected. Future studies could
investigate how varying the complexity of spontaneous speech tasks influences the interaction
between the different DMN subsystems and control networks such as the MDN and SCN.
Tasks designed to elicit more unusual speech content may preferentially engage the dmDMN
and SCN, reflecting the need for more controlled retrieval and selection during creative idea
generation. This work could deepen our understanding about the role of the DMN in automatic
and controlled processing, as well as the coordination between these networks in supporting
divergence in speech.

156
 In summary, our current work explored the relationship between coherence and
creativity in spontaneous speech, revealing a trade-off between maintaining logical flow and
introducing semantically diverse ideas. We demonstrated that speech coherence is supported
by the Multiple Demand Network, which facilitates the organisation and management of
speech content; while divergence involved the interaction between the Default Mode and
Multiple Demand Networks. These findings contribute to our understanding of how we
balanced structured communication with creative expression in everyday speech, and the
neurocognitive networks that support these processes. Future work could further explore how
task complexity influences the engagement of these networks.

157
 Chapter 6

General Discussion
Creativity, the ability to generate novel and useful ideas, is a dynamic and complex
process that draws from a diverse toolbox of cognitive resources. The semantic system is
essential for this process: semantic knowledge provides the foundation, serving as a rich source
of concepts and associations that can be used to form new ideas (Abraham & Bubic, 2015;
Benedek et al., 2023; Gerver et al., 2023). However, using this knowledge effectively requires
control over how information is retrieved and selected. For this, we rely on semantic control
mechanisms that allow us to efficiently navigate our stored knowledge, inhibiting irrelevant or
dominant associations, and strategically selecting and combining elements in unconventional
ways. Thus, examining how semantic control works to guide and shape creative thought is
essential to understanding the cognitive processes that underlie our ability to think creatively.
Despite the potential importance of these processes, the specific contributions of
semantic control mechanisms to creative thought have not been sufficiently explored. To fill
this gap in the literature, the aim of this thesis was to investigate how different aspects of
semantic control influence the ability to generate and evaluate creative ideas. This thesis
addressed these aims through a series of experimental studies, each focusing on different facets
of creative thinking. First, Chapter 3 examined how individual differences in semantic
knowledge and control abilities, over and above domain-general executive functions and fluid
intelligence, contribute to divergent and convergent thinking. Further, the chapter explored
whether these age-related differences in these abilities influence creative thought. Second,
Chapter 4 investigated how the ability to generate and evaluate ideas in problems with
multiple constraints is related to divergent and convergent thinking, and further, the effects of
individual differences in search and retrieval dynamics. Finally, Chapter 5 explored the
behavioural and neural mechanisms underlying creativity and coherence in spontaneous
speech, providing insights into how these processes unfold in more naturalistic contexts. In this
discussion chapter, I will first present a summary of the main findings from each chapter, and
then integrate these findings into a broader understanding of the role of semantic control in
creative thinking.

6.1. Chapter Summaries

Chapter 3 supports the idea that divergent and convergent thinking draw on a distinct
set of cognitive abilities, with differential recruitment of these abilities across the lifespan.

158
Although older adults exhibited declines in domain-general executive functions and fluid
intelligence, they possessed more semantic knowledge and better controlled retrieval abilities
than their younger counterparts, and showed superior performance on convergent thinking
tasks, like the Remote Associates Test (RAT). Semantic knowledge and updating (i.e., the
executive ability to hold and manipulate the contents of working memory) were closely tied to
convergent thinking across both age groups. This implies that the contents of semantic memory
and the ability to update information in an online fashion are critical tools drawn upon by both
groups. Interestingly, divergent thinking abilities, measured by the Alternate Uses Task (AUT)
were similar across younger and older adults, although the two groups appeared to rely on
different cognitive strategies to achieve creative outcomes. My findings confirmed that
semantic control mechanisms play a significant role in divergent thinking, particularly in
younger adults. This distinction implies age-related differences in recruitment of cognitive
abilities in the pursuit of divergent, but not convergent, thinking. Younger adults relied more
on their semantic control mechanisms to generate novel ideas, while older adults depended on
domain-general executive mechanisms, such as inhibition and shifting. Moreover, older adults
may be drawing on their stored knowledge to think creatively, aligning with the default-
executive coupling hypothesis of aging (DECHA; Spreng & Turner, 2019) which posits that
older adults compensate for their declines in cognitive control by leveraging their extensive
semantic knowledge and life experience. This adaptability may help preserve their creative
abilities, particularly in tasks requiring the flexible use of knowledge.
Chapter 4 indicated that the ability to generate candidate responses under multiple
constraints is related to fluency in divergent thinking tasks, such as the AUT. This aligns with
previous work indicating that broad retrieval ability contributes to the quantity of ideas
produced in creative tasks (Forthmann, Jendryczko, et al., 2019; Miroshnik et al., 2023).
However, the ability to evaluate and select candidate ideas was negatively associated with the
ability to produce unique responses to AUT prompts, suggesting that the evaluative skills
required in multiply-constrained problem-solving tasks may be different to those required for
divergent thinking. Furthermore, more effective semantic search was found to involve a
balance between the formation of multiple clusters (clustering breadth) and the thorough
exploration within a cluster (clustering depth). The interaction between these two strategies
predicted success on both the RAT and the AUT, highlighting the different routes to creative
success. Specifically, a broad search strategy traversing a larger amount of the semantic space
is generally advantageous, but, deep focus within a single cluster may also be beneficial in
certain contexts, particularly when fewer clusters are formed overall. This balance between

159
exploration and exploitation aligns with optimal foraging theories, suggesting that the most
effective search strategy may vary depending on task demands and individual differences
(Hills, Todd, & Jones, 2015; Todd & Hills, 2020). Importantly, clustering and switching may
not be fixed traits but may be adaptive strategies that individuals flexibly employ, depending
on the contextual demands of the task (Mekern et al., 2019). The effectiveness of these
strategies can vary significantly across different types of creative tasks: for example, on the
RAT, which involves finding new links between remote concepts, a strategy that balances
clustering and switching can be beneficial. These findings underscore the importance of
considering both quantity and depth of clusters formed during semantic search, highlighting
the complex interplay between exploitation and exploration in creative thought. This nuanced
understanding of search strategies provides insights into the multiple pathways that can lead to
success in creative problem-solving tasks.
Chapter 5 explored the complex relationship between coherence and creativity in
spontaneous speech, and the neural mechanisms supporting these speech characteristics. First,
the study identified a negative relationship between global coherence and divergence in
naturalistic speech, indicating that well-structured and logical speech often comes at the
expense of originality and diversity of ideas. Moreover, exploratory analyses indicated that
speakers naturally modulate their speech based on conversational goals, such as producing
more divergent but less coherent speech when asked to be entertaining. This suggests a
fundamental trade-off between coherence and creativity. At the neural level, this study was the
first to establish a link between coherence in speech and the Multiple Demand Network
(MDN), a set of regions involved in managing complex cognitive processes. Here, the MDN
may organize and manage the logical flow of speech, ensuring ideas are relevant and follow a
logical flow. In contrast, divergence in speech was related to increased functional connectivity
between the MDN and the Default Mode Network (DMN), a neural system associated with
internally directed thought processes such as constructive daydreaming and mental time travel
(e.g., Andrews-Hanna, 2012; Andrews-Hanna et al., 2018). This default-executive coupling
reflects the need for both generative processes, supported by the DMN, and organizational
control, supported by the MDN, in generating more divergent speech content. Further, this is
consistent with prior research indicating that creativity relies on the dynamic interplay between
associative and executive processes (Beaty et al., 2015, 2016). Contrary to expectations, the
Semantic Control Network (SCN) was not significantly implicated in either coherence or
divergence of speech. This may be due to the nature of the task, where participants may have
relied more on automatic retrieval from of well-established knowledge or episodic memory,

160
reducing the demand for semantic control processes. These findings suggest that balancing
structured communication with creative expression requires dynamic interactions between the
DMN and the MDN, with implications for how task requirements may differentially engage
these networks, depending on their complexity and novelty.

6.2. Broader Theoretical Implications

6.2.1. Controlled Semantic Cognition

The Controlled Semantic Cognition (CSC; Lambon Ralph et al., 2017) framework
proposes that semantic cognition is underpinned by two interacting systems of representation
and control. The results of Chapter 3 and Chapter 4 situate creative thinking within the CSC
framework, furthering our understanding of the contributions of semantic control mechanisms
to more complex cognitive operations. While much of the existing creativity literature has
focused on the content, structure and organization of semantic memory, the current work
highlights the importance of control mechanisms in governing how this knowledge is accessed
and utilized. Rich semantic memory structures, i.e., densely connected concepts, can both
facilitate and constrain creative thinking (Beaty et al., 2023). On one hand, they provide a rich
network of associations to draw from, allowing for the generation of a large quantity of ideas.
On the other hand, they increase the risk of ‘fixation’ on conventional ideas due to interference
from strong, well-established associations that constrain the ability to think beyond familiar
conceptual links. Prior research has suggested that these constraints can be mitigated by
domain-general processes like executive functions, fluid intelligence and broad retrieval
abilities (Beaty, Benedek, et al., 2014, 2014; Beaty & Silvia, 2012, 2012, 2013; Benedek,
Franz, et al., 2012). However, little attention has been devoted to understanding how specific
semantic control processes contribute to creative thought. Thus, the current findings address
this gap by highlighting the critical role of control processes within creativity.
Specifically, during creative thinking, controlled retrieval processes enable individuals
to flexibly manipulate activation within the semantic network, suppressing dominant ideas in
favor of goal-directed search for more remote, but relevant, associations. This search activates
multiple different associations, prompting semantic selection mechanisms that allow
individuals to evaluate and resolve competition between ideas in line with task objectives. The
theoretical and neural implications of these findings for creativity research will be explored in
the following sections.

161
6.2.2. Dual Process Theories
According to dual-process models, creativity emerges from an interplay between
bottom-up associative processes and top-down controlled processes (Barr et al., 2015;
Kleinmintz et al., 2019; Nijstad et al., 2010; Sowden et al., 2015). Spontaneous, associative
processes provide the raw materials for creative ideas and are thought to operate primarily
during the generation phase of creative thinking. In contrast, top-down executive control is
required during the evaluation phase to refine and test creative ideas in line with task objectives.
The results from Chapter 3 and Chapter 4 support and extend these dual-process models,
highlighting the importance of semantic control mechanisms to creative thinking beyond the
contributions of broad executive functions. Specifically, controlled retrieval and selection
processes may help to shape retrieval and select ideas in line with task constraints of novelty
and usefulness. This particularly useful in tasks such as the AUT and RAT, where retrieval
flexibility and inhibition of stereotypical responses is critical.
Creative thinking is often associated with broad retrieval ability (Forthmann,
Jendryczko, et al., 2019; Miroshnik et al., 2023). Broad retrieval is a higher order factor in the
Cattell-Horn-Carroll model of intelligence (Cattell, 1971), reflecting the ability to fluently
retrieve information from memory (Schneider & McGrew, 2018). However, let us consider the
measures used to index these two abilities. Broad retrieval ability is commonly assessed using
fluency tasks, which prioritize the quantity of ideas generated within a limited time (‘Generate
as many animal names as you can in 60 seconds’). Conversely, controlled retrieval tasks (e.g.,
Hoffman, 2018) present a different challenge: the retrieval of specific, weakly-related targets
(ring) that are not automatically activated by a given probe (iron). Here, individuals must
engage in an active search and selection process, while inhibiting irrelevant responses, to
extract this specific information. Thus, while fluency tasks emphasize the retrieval of numerous
concepts in response to a broad cue—where the specific content is less important if it fits the
general category—controlled retrieval tasks demand the identification of a specific piece of
semantic information that links two concepts, akin to tasks like the RAT. This distinction
highlights how cognitive processes may be differentially recruited for each task, with fluency
focusing on the breadth of retrieval, and controlled retrieval emphasizing depth and specificity
in the search for relevant information.
Chapter 4 demonstrated that the ability to generate ideas to multiply constrained
problems is linked to creative fluency, i.e., the number of ideas produced on the AUT. This
finding aligns with previous research showing a strong connection between broad retrieval and

162
creative fluency, suggesting they may share a common cognitive basis, and fall within a broader
construct indexing idea generation (Forthmann, Jendryczko, et al., 2019; Miroshnik et al.,
2023; Weiss et al., 2024). However, recent work suggests that the relationship between creative
fluency and originality of ideas diminishes when accounting for crystallized intelligence and
working memory factors (Weiss et al., 2024). This indicates that while the ability to fluently
generate multiple ideas is valuable, it does not necessarily translate into the generation of
original ideas. My study in Chapter 4 corroborates this, showing no relationship between
generative abilities and the creative quality of AUT responses.
Furthermore, cognitive processes may be differentially recruited for various fluency
tasks. For instance, category fluency relies more on free association and automatic retrieval
driven by the taxonomic organization of semantic information; whereas letter fluency imposes
higher executive demands, requiring the suppression of automatic semantic associations and
focus on phonological or orthographic constraints (Marko et al., 2022). Additionally, tasks
requiring dissociative retrieval (i.e., generating unrelated words) share similarities with letter
fluency tasks, as both tasks likely involve inhibitory control processes. Thus, while creative
fluency is closely tied to a general retrieval ability, tasks with higher executive demands may
require additional control. This suggests that fluency and goal-directed retrieval are not entirely
independent constructs; but, both are required for effective creative thinking, especially when
executive demands are higher.
These insights suggest that controlled processes are not only crucial during the
evaluation phase, but may also play a significant role in the generation phase of creative
thinking (Sowden et al., 2015). This challenges the idea that associative processes alone drive
idea generation and suggests that controlled processes might be involved in actively guiding
the search for relevant and novel ideas. In this context, the dynamics of strategic search become
particularly relevant.

6.2.3. Strategic Search

This brings us to a key aspect of creative thinking: the strategies used during the search
for ideas, which was a central focus in Chapter 4. Two search components: clustering
(generating ideas within a category) and switching (shifting to new categories) have been
identified in verbal fluency tasks (e.g., Troyer, 2000; Troyer et al., 1997) and in the creative
process (Mastria et al., 2021; Ovando-Tellez et al., 2022). Chapter 4 extends this
understanding, underscoring the importance of semantic search dynamics in the context of
efficient goal-directed retrieval under multiple cue constraints. Two search components were

163
identified: clustering depth, staying within a semantic category; and clustering breadth,
moving between semantic categories. Clustering depth was interpreted as representing an
automatic, associative processes; while clustering breadth was taken to reflect controlled,
evaluative processes that guide search. Crucially, creative performance relied on a complex
interplay between these two search components, highlighting the necessity for controlled
processes not only in the evaluation of ideas, but also within the generation phase itself.
This notion corresponds to the dual-pathway model (Nijstad et al., 2010), which
outlines two pathways to creativity, both of which involve deliberative thinking. The flexibility
pathway, characterized by the generation of a wide variety of ideas across different categories,
involves both associative thinking and the ability to shift between cognitive categories. On the
other hand, the persistence pathway emphasizes the systematic and effortful exploration within
a semantic category, allowing for the refinement of ideas. While both pathways contribute to
creativity, the flexibility pathway is more strongly related to creative performance. Importantly,
creative success depends on the ability to adeptly shift between the two pathways, generating
a breadth of ideas while focusing deeply on refining a select few. The dual-pathway model is
one of many in a long tradition of dual-process models of creativity (Sowden et al., 2015),
mapping onto processes of idea generation and evaluation, where flexibility supports the
generation of a broad range of ideas, and persistence ensures their systematic testing and
refinement.

6.2.4. Extending the Dual Process: MemiC Framework

Building on these foundations, future work could investigate the specific role of
semantic control mechanisms within the Memory in Creative Ideation (MemiC; Benedek et al.,
2023) framework. The MemiC framework breaks down the generation and evaluation phases
of creativity into distinct substages. During the generation process, it distinguishes between a
search phase, where pertinent information is retrieved from memory in a goal-directed manner;
and a construction phase, where this information is meaningfully recombined to suit task
objectives.
My findings suggest that semantic control mechanisms are integral to these substages.
In the search phase, broad retrieval ability facilitates the production of multiple possible
solutions, while controlled retrieval abilities enable a goal-directed search, by inhibiting
irrelevant associations and selectively retrieving pertinent information. This is especially
relevant for tasks like the AUT or RAT, where the challenge lies in extracting specific
information from stored knowledge. During the construction phase, new connections are forged

164
between ideas. Here, controlled retrieval continues to play a critical role in the conscious and
effortful search for relevant associations amongst retrieved words. For example, take the RAT
problem ‘cottage – blue – cake’. Considering the cue word ‘cake’ may prompt individuals to
think of types of cake (e.g., chocolate cake, cheesecake) which could then be strategically
combined with ‘blue’ (as in blue cheese) and ‘cottage’ (as in cottage cheese). Here, semantic
selection mechanisms are also essential, as individuals must determine which retrieved
elements should be combined or discarded.
On the other hand, the evaluation phase involves the assessment of candidate ideas for
both novelty and effectiveness. Assessing novelty involves the comparison of candidate ideas
with stored knowledge, evaluating ideas that cannot be retrieved from memory as being more
original. For instance, in the context of the AUT, using a belt as a sushi plate can be classified
as creative, as this use has not been encountered before. Similarly, the effectiveness of an idea
is assessed by comparing it with existing semantic information, such as object properties and
affordances, or previous encounters with the object.

6.2.5. Key Distinction between New and Old Ideas: The Episodic vs Semantic Question
The MemiC framework emphasizes the contributions of both semantic and episodic
memory to creative thinking. While episodic memory does not fall within the purview of this
thesis, it is important to take into account how these two memory systems influence the
generation of creative ideas. Historically, the literature has indicated the existence of two
functionally distinct memory systems underlying long-term memory (Tulving, 2002). While
semantic memory refers to conceptual knowledge about the world, episodic memory represents
specific events that we have experienced through over our lifespan. More recent work has
challenged this classical dichotomy, suggesting these systems may not be as distinct as
previously thought (Irish & Vatansever, 2020). For example, the controlled retrieval of both
weakly-encoded information from semantic and episodic memory may rely on shared neural
resources, specifically, the LIFG (Vatansever et al., 2021). This finding suggests that the level
of control required for memory retrieval, rather than the memory type itself, may play a pivotal
role in shaping how memories are processed in the brain.
There is a growing body of evidence highlighting the link between episodic memory
and divergent thinking. For example, participants report using episodic memory strategies,
especially early in the generation process, to generate novel uses for an object – but, notably,
these strategies do not produce responses that are novel to the individual (Gilhooly et al., 2007).
In addition, episodic induction - training in recollection of past details to facilitate episodic

165
retrieval - has been found to enhance fluency and flexibility of responses on the AUT in both
younger (Madore et al., 2015) and older people (Madore et al., 2016). However, these findings
were recently contested by Ahmed et al. (2023), who included a no-induction baseline and a
control group, and found no significant effects of episodic retrieval on divergent thinking, with
older adults simply performing better than younger adults. Together, while these results suggest
that the episodic induction may not bestow any advantages or disadvantages on divergent
thinking performance, they do point to the involvement of episodic memory. Moreover,
neuroimaging studies have indicated that episodic retrieval and divergent thinking may activate
overlapping brain areas (Beaty et al., 2020; Beaty, Thakral, et al., 2018), particularly within the
default mode network, which supports both episodic and semantic processing (Kim, 2016).
Specifically, creating connections based on strong associations may rely more heavily on the
episodic system as people may be drawing from their store of personal memories (Krieger-
Redwood et al., 2023).
In Chapter 3, I observed that semantic control mechanisms are particularly important
for divergent thinking in the young adults. This observation raises the question of why these
mechanisms might be less critical for older adults. Considering the nature of AUT provides
some insights: this task often involves the integration of semantic and episodic information, as
individuals draw on their knowledge of object properties as well as personal experiences with
these objects. With age, there is a ‘semanticization’ of episodic memory (Spreng & Turner,
2019), i.e., detailed specific memories of personal events gradually become more abstract and
integrated into the broader semantic knowledge. This process of semanticization can benefit
older adults in tasks like the AUT: to effectively perform the task, they may be leveraging their
wealth of personal experiences that have become interconnected with semantic knowledge, in
lieu of relying on semantic control. Interestingly, this notion of compensatory mechanisms is
further supported by findings indicating stronger default-executive network coupling in older
adults during creative tasks, compared to their younger counterparts (Adnan, Beaty, Lam, et
al., 2019; Adnan, Beaty, Silvia, et al., 2019). This stronger coupling may enhance the older
adults’ abilities to engage in creative cognition by facilitating the integration of stored
knowledge and executive processes. This can be interpreted in support of the idea of that older
adults rely on automatic retrieval of strong semantic associations and episodic memories in
their creative thinking, thereby reducing the necessity for Semantic Control Network
engagement.
This observation connects to the findings from Chapter 5, where default-executive
coupling, rather than the SCN, was related to divergence in spontaneous speech. The reduced

166
reliance on the SCN could be attributed to a dependence on automatic retrieval from episodic
memory. Indeed, the nature of the prompts, which required participants to speak about familiar
contexts, likely involved strong episodic associations, thereby reducing the need for active
semantic control. This aligns with existing research suggesting that the DMN and SCN may
facilitate associative thinking through different memory systems (Krieger-Redwood et al.,
2023). Specifically, the DMN may support the retrieval of strong, well-established knowledge
or episodic memories, while the SCN may be integral when new conceptual links need to be
forged between concepts, especially in contexts where past experiences cannot be relied on.
Together, the observations suggest a complex interplay between different memory systems and
neural networks in creative thought.

6.2.6. Segmenting the Gyrus: The Distinct Roles of BA 45 and BA 47 in Semantic Control
The left inferior frontal gyrus (LIFG) is a key player in language processing, memory
retrieval, and cognitive control. Importantly, this region is not functionally homogenous. It
comprises of distinct subdivisions – BA 44 (pars opercularis), BA 45 (pars triangularis) and
BA 47 (pars orbitalis) – each uniquely contributing to aspects of cognitive processing (Badre
et al., 2005; Fedorenko & Blank, 2020). Of particular interest to the current work are BA 45
and BA 47, both closely linked to semantic control. Specifically, mid-to-anterior portions of
the LIFG (i.e., BA 47/pars orbitalis) have been linked to a specialized role in the controlled
retrieval of weak semantic information and episodic information (Badre et al., 2005; Barredo
et al., 2016), preferentially activating in response to semantic control demands rather than non-
semantic tasks (Hodgson et al., 2021). In contrast, mid-to-posterior regions of the LIFG (i.e.,
BA 45/pars triangularis) have been linked to post-retrieval semantic selection processes,
particularly in resolving competition between candidate ideas (Badre et al., 2005). BA 45
demonstrates more domain-generality than BA 47, supporting competition resolution in
semantic, episodic and working memory domains (Badre & Wagner, 2002, 2005; Branzi &
Lambon Ralph, 2023; Dobbins & Wagner, 2005). The specialized roles of the LIFG
subdivisions correspond to their distinct connectivity patterns: BA 45 is connected to regions
of the MDN, like the DLPFC (Jung et al., 2022); while BA 47 is strongly connected to the
ATLs, the representational hub of the semantic system (Binney et al., 2012) and does not
overlap with the MDN (Assem et al., 2020; Duncan, 2010; Fedorenko et al., 2013). These
connectivity patterns emphasise the functional specialization of these regions in managing
different aspects of cognitive and semantic control.

167
 Though the LIFG has been consistently implicated in creative thinking, the distinct
contributions of BA 45 and BA 47 to creative cognition have not yet been clarified. Given the
functional diversity of this region, future work should explore how subdivisions of the LIFG
differentially contribute to creative thought. This could provide valuable insights into the neural
underpinnings of creativity, specifically in understanding how controlled semantic cognition
shapes creative thinking.

6.2.7. Bringing Creativity into the Real World

The field of creative cognition research has been predominantly shaped by the use of
divergent thinking tasks, such as the AUT, which focus on assessing an individual’s ability to
generate a large quantity of ideas that are both novel and appropriate. The use of these tasks
has provided invaluable insights into cognitive mechanisms underlying creative processing,
both at the word (e.g., Beaty & Kenett, 2023; Beaty & Silvia, 2012; Benedek et al., 2012;
Ovando-Tellez et al., 2022) and sentence level (Weinstein et al., 2022). However, these tasks
fail to capture the full scope of creative thinking as it occurs in more complex contexts. In the
real-world, creativity is rarely restricted to idea generation within a vacuum – instead, it often
involves the ability to tie ideas together in coherent narratives or adapt to diverse contexts.
Recognizing this, recent studies have expanded the assessment of creativity into more complex
domains like creative and academic writing, story generation and poetry (Bower et al., 2023;
D’Souza, 2021b; Fan et al., 2023; R. He et al., 2022; Liu et al., 2015; Rastelli et al., 2024;
Zedelius et al., 2019). Chapter 5 adds to this literature by investigating creativity in a more
ecologically valid context: spontaneous narrative speech, a dynamic form of expression that
closely mirrors everyday creative activity. Crucially, the study presented in Chapter 5
identified a pattern of default-executive coupling, commonly observed in divergent thinking,
underlying the spontaneous generation of narrative speech. This discovery is noteworthy, as it
broadens our understanding of creative cognition by revealing that the same cognitive
processes involved in intentional creative acts also contribute to naturalistic, unstructured
creative acts. By focusing on extemporaneous speech, this work highlights the importance of
investigating creativity not only when people are explicitly prompted to generate ideas, but
also when creativity emerges organically in conversation.

6.3. Conclusion
Overall, this thesis explored whether specific aspects of semantic control contribute to
creative thinking. In line with the Controlled Semantic Cognition framework and dual-process
models of creativity, the findings presented across this thesis suggest that creativity emerges as

168
a product of spontaneous associative processes and controlled executive processes.
Importantly, the results of this thesis suggest that semantic control mechanisms may be
particularly important for guiding the search and refinement of ideas to meet specific
constraints. Further, it highlights the necessity of controlled processing during both idea
generation and evaluation, adding an important dimension to dual-process models of creativity.
Thus, the relationship between associative and controlled thinking may not just be crucial for
judging ideas, but also for efficient search and retrieval. As research continues to explore the
cognitive processes underlying creativity, it is essential to further investigate the nuanced
contributions of the semantic control system both at the behavioural and neural levels.

169
 References
AbdulSabur, N. Y., Xu, Y., Liu, S., Chow, H. M., Baxter, M., Carson, J., & Braun, A. R.

(2014). Neural correlates and network connectivity underlying narrative production

and comprehension: A combined fMRI and PET study. Cortex, 57, 107–127.

[Link]

Abraham, A. (2014). Creative thinking as orchestrated by semantic processing vs. Cognitive

control brain networks. Frontiers in Human Neuroscience, 8.

[Link]

Abraham, A., Beudt, S., Ott, D. V. M., & Yves Von Cramon, D. (2012). Creative cognition

and the brain: Dissociations between frontal, parietal–temporal and basal ganglia

groups. Brain Research, 1482, 55–70. [Link]

Abraham, A., & Bubic, A. (2015). Semantic memory as the root of imagination. Frontiers in

Psychology, 6. [Link]

Acar, S., Ogurlu, U., & Zorychta, A. (2023). Exploration of discriminant validity in divergent

thinking tasks: A meta-analysis. Psychology of Aesthetics, Creativity, and the Arts,

17(6), 705–724. [Link]

Adank, P. (2012). The neural bases of difficult speech comprehension and speech production:

Two Activation Likelihood Estimation (ALE) meta-analyses. Brain and Language,

122(1), 42–54. [Link]

Adnan, A., Beaty, R., Lam, J., Spreng, R. N., & Turner, G. R. (2019). Intrinsic default—

Executive coupling of the creative aging brain. Social Cognitive and Affective

Neuroscience, 14(3), 291–303. [Link]

Adnan, A., Beaty, R., Silvia, P., Spreng, R. N., & Turner, G. R. (2019). Creative aging:

Functional brain networks associated with divergent thinking in older and younger

170
 adults. Neurobiology of Aging, 75, 150–158.

[Link]

Adólfsdóttir, S., Wollschlaeger, D., Wehling, E., & Lundervold, A. J. (2017). Inhibition and

Switching in Healthy Aging: A Longitudinal Study. Journal of the International

Neuropsychological Society, 23(1), 90–97.

[Link]

Ahmed, H., Pauly-Takacs, K., & Abraham, A. (2023). Evaluating the effects of episodic and

semantic memory induction procedures on divergent thinking in younger and older

adults. PLOS ONE, 18(6), e0286305. [Link]

Andrews-Hanna, J. R. (2012). The Brain’s Default Network and Its Adaptive Role in Internal

Mentation. The Neuroscientist, 18(3), 251–270.

[Link]

Andrews-Hanna, J. R., Irving, Z. C., Fox, K. C. R., Spreng, R. N., & Christoff, K. (2018).

The Neuroscience of Spontaneous Thought (K. Christoff & K. C. R. Fox, Eds.; Vol.

1). Oxford University Press. [Link]

Andrews‐Hanna, J. R., Smallwood, J., & Spreng, R. N. (2014). The default network and self‐

generated thought: Component processes, dynamic control, and clinical relevance.

Annals of the New York Academy of Sciences, 1316(1), 29–52.

[Link]

Arbuckle, T. Y., & Gold, D. P. (1993). Aging, inhibition, and verbosity. Journal of

Gerontology, 48(5), P225–P232.

Argiris, G., MacPherson, S. E., Della Sala, S., & Foley, J. A. (2020). The relationship

between dual-tasking and processing speed in healthy aging. Psychology &

Neuroscience, 13(3), 375–389. [Link]

171
Ashburner, J. (2007). A fast diffeomorphic image registration algorithm. Neuroimage, 38(1),

95–113.

Assem, M., Glasser, M. F., Van Essen, D. C., & Duncan, J. (2020). A Domain-General

Cognitive Core Defined in Multimodally Parcellated Human Cortex. Cerebral Cortex,

30(8), 4361–4380. [Link]

Awad, M., Warren, J. E., Scott, S. K., Turkheimer, F. E., & Wise, R. J. S. (2007). A Common

System for the Comprehension and Production of Narrative Speech. The Journal of

Neuroscience, 27(43), 11455–11464. [Link]

06.2007

Baddeley, A. D., Emslie, H., & Nimmo-Smith, I. (1992). The speed and capacity of

language-processing test. Thames Valley Test Company.

Baddeley, A., Emslie, H., & Nimmo‐Smith, I. (1993). The Spot‐the‐Word test: A robust

estimate of verbal intelligence based on lexical decision. British Journal of Clinical

Psychology, 32(1), 55–65. [Link]

Badre, D., Poldrack, R. A., Paré-Blagoev, E. J., Insler, R. Z., & Wagner, A. D. (2005).

Dissociable Controlled Retrieval and Generalized Selection Mechanisms in

Ventrolateral Prefrontal Cortex. Neuron, 47(6), 907–918.

[Link]

Badre, D., & Wagner, A. D. (2002). Semantic retrieval, mnemonic control, and prefrontal

cortex. Behavioral and Cognitive Neuroscience Reviews, 1(3), 206–218.

Badre, D., & Wagner, A. D. (2005). Frontal Lobe Mechanisms that Resolve Proactive

Interference. Cerebral Cortex, 15(12), 2003–2012.

[Link]

172
Badre, D., & Wagner, A. D. (2007). Left ventrolateral prefrontal cortex and the cognitive

control of memory. Neuropsychologia, 45(13), 2883–2901.

[Link]

Barbot, B. (2018). The Dynamics of Creative Ideation: Introducing a New Assessment

Paradigm. Frontiers in Psychology, 9, 2529.

[Link]

Barker, M. S., Young, B., & Robinson, G. A. (2017). Cohesive and coherent connected

speech deficits in mild stroke. Brain and Language, 168, 23–36.

[Link]

Barr, N., Pennycook, G., Stolz, J. A., & Fugelsang, J. A. (2015). Reasoned connections: A

dual-process perspective on creative thought. Thinking & Reasoning, 21(1), 61–75.

[Link]

Barredo, J., Verstynen, T. D., & Badre, D. (2016). Organization of cortico-cortical pathways

supporting memory retrieval across subregions of the left ventrolateral prefrontal

cortex. Journal of Neurophysiology, 116(3), 920–937.

[Link]

Barron, F. (1955). The disposition toward originality. The Journal of Abnormal and Social

Psychology, 51(3), 478.

Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models

Using lme4. Journal of Statistical Software, 67(1), 1–48.

[Link]

Batey, M., Chamorro-Premuzic, T., & Furnham, A. (2009). Intelligence and personality as

predictors of divergent thinking: The role of general, fluid and crystallised

intelligence. Thinking Skills and Creativity, 4(1), 60–69.

[Link]

173
Batey, M., & Furnham, A. (2006). Creativity, intelligence, and personality: A critical review

of the scattered literature. Genetic, Social, and General Psychology Monographs, 132,

355–429. [Link]

Beaty, R. E., Benedek, M., Barry Kaufman, S., & Silvia, P. J. (2015). Default and Executive

Network Coupling Supports Creative Idea Production. Scientific Reports, 5(1), 10964.

[Link]

Beaty, R. E., Benedek, M., Silvia, P. J., & Schacter, D. L. (2016). Creative Cognition and

Brain Network Dynamics. Trends in Cognitive Sciences, 20(2), 87–95.

[Link]

Beaty, R. E., Benedek, M., Wilkins, R. W., Jauk, E., Fink, A., Silvia, P. J., Hodges, D. A.,

Koschutnig, K., & Neubauer, A. C. (2014). Creativity and the default network: A

functional connectivity analysis of the creative brain at rest. Neuropsychologia, 64,

92–98. [Link]

Beaty, R. E., Chen, Q., Christensen, A. P., Kenett, Y. N., Silvia, P. J., Benedek, M., &

Schacter, D. L. (2020). Default network contributions to episodic and semantic

processing during divergent creative thinking: A representational similarity analysis.

NeuroImage, 209, 116499. [Link]

Beaty, R. E., Christensen, A. P., Benedek, M., Silvia, P. J., & Schacter, D. L. (2017). Creative

constraints: Brain activity and network dynamics underlying semantic interference

during idea production. NeuroImage, 148, 189–196.

[Link]

Beaty, R. E., & Johnson, D. R. (2021). Automating creativity assessment with SemDis: An

open platform for computing semantic distance. Behavior Research Methods, 53(2),

757–780. [Link]

174
Beaty, R. E., Johnson, D. R., Zeitlen, D. C., & Forthmann, B. (2022). Semantic Distance and

the Alternate Uses Task: Recommendations for Reliable Automated Assessment of

Originality. Creativity Research Journal, 34(3), 245–260.

[Link]

Beaty, R. E., & Kenett, Y. N. (2023). Associative thinking at the core of creativity. Trends in

Cognitive Sciences, 27(7), 671–683. [Link]

Beaty, R. E., Kenett, Y. N., Christensen, A. P., Rosenberg, M. D., Benedek, M., Chen, Q.,

Fink, A., Qiu, J., Kwapil, T. R., Kane, M. J., & Silvia, P. J. (2018). Robust prediction

of individual creative ability from brain functional connectivity. Proceedings of the

National Academy of Sciences, 115(5), 1087–1092.

[Link]

Beaty, R. E., Kenett, Y. N., Hass, R. W., & Schacter, D. L. (2023). Semantic memory and

creativity: The costs and benefits of semantic memory structure in generating original

ideas. Thinking & Reasoning, 29(2), 305–339.

[Link]

Beaty, R. E., Seli, P., & Schacter, D. L. (2019). Network neuroscience of creative cognition:

Mapping cognitive mechanisms and individual differences in the creative brain.

Current Opinion in Behavioral Sciences, 27, 22–30.

[Link]

Beaty, R. E., & Silvia, P. J. (2012). Why do ideas get more creative across time? An executive

interpretation of the serial order effect in divergent thinking tasks. Psychology of

Aesthetics, Creativity, and the Arts, 6(4), 309–319. [Link]

Beaty, R. E., & Silvia, P. J. (2013). Metaphorically speaking: Cognitive abilities and the

production of figurative language. Memory & Cognition, 41(2), 255–267.

[Link]

175
Beaty, R. E., Silvia, P. J., Nusbaum, E. C., Jauk, E., & Benedek, M. (2014). The roles of

associative and executive processes in creative cognition. Memory & Cognition,

42(7), 1186–1197. [Link]

Beaty, R. E., Thakral, P. P., Madore, K. P., Benedek, M., & Schacter, D. L. (2018). Core

Network Contributions to Remembering the Past, Imagining the Future, and Thinking

Creatively. Journal of Cognitive Neuroscience, 30(12), 1939–1951.

[Link]

Beaty, R. E., Zeitlen, D. C., Baker, B. S., & Kenett, Y. N. (2021). Forward flow and creative

thought: Assessing associative cognition and its role in divergent thinking. Thinking

Skills and Creativity, 41, 100859. [Link]

Becker, M., Sommer, T., & Kühn, S. (2020). Inferior frontal gyrus involvement during search

and solution in verbal creative problem solving: A parametric fMRI study.

NeuroImage, 206, 116294. [Link]

Bendetowicz, D., Urbanski, M., Garcin, B., Foulon, C., Levy, R., Bréchemier, M.-L., Rosso,

C., Thiebaut De Schotten, M., & Volle, E. (2018). Two critical brain networks for

generation and combination of remote associations. Brain, 141(1), 217–233.

[Link]

Benedek, M. (2024). On the relationship between creative potential and creative

achievement: Challenges and future directions. Learning and Individual Differences,

110, 102424. [Link]

Benedek, M., Beaty, R. E., Schacter, D. L., & Kenett, Y. N. (2023). The role of memory in

creative ideation. Nature Reviews Psychology, 2(4), 246–257.

[Link]

Benedek, M., Beaty, R., Jauk, E., Koschutnig, K., Fink, A., Silvia, P. J., Dunst, B., &

Neubauer, A. C. (2014). Creating metaphors: The neural basis of figurative language

176
 production. NeuroImage, 90, 99–106.

[Link]

Benedek, M., & Fink, A. (2019). Toward a neurocognitive framework of creative cognition:

The role of memory, attention, and cognitive control. Current Opinion in Behavioral

Sciences, 27, 116–122. [Link]

Benedek, M., Franz, F., Heene, M., & Neubauer, A. C. (2012). Differential effects of

cognitive inhibition and intelligence on creativity. Personality and Individual

Differences, 53(4), 480–485. [Link]

Benedek, M., & Jauk, E. (2018). Spontaneous and Controlled Processes in Creative

Cognition (K. Christoff & K. C. R. Fox, Eds.; Vol. 1). Oxford University Press.

[Link]

Benedek, M., Jauk, E., Sommer, M., Arendasy, M., & Neubauer, A. C. (2014). Intelligence,

creativity, and cognitive control: The common and differential involvement of

executive functions in intelligence and creativity. Intelligence, 46, 73–83.

[Link]

Benedek, M., Kenett, Y. N., Umdasch, K., Anaki, D., Faust, M., & Neubauer, A. C. (2017).

How semantic memory structure and intelligence contribute to creative thought: A

network science approach. Thinking & Reasoning, 23(2), 158–183.

[Link]

Benedek, M., Könen, T., & Neubauer, A. C. (2012). Associative abilities underlying

creativity. Psychology of Aesthetics, Creativity, and the Arts, 6(3), 273–281.

[Link]

Benedek, M., & Neubauer, A. C. (2013). Revisiting Mednick’s Model on Creativity‐Related

Differences in Associative Hierarchies. Evidence for a Common Path to Uncommon

177
 Thought. The Journal of Creative Behavior, 47(4), 273–289.

[Link]

Binder, J. R., & Desai, R. H. (2011). The neurobiology of semantic memory. Trends in

Cognitive Sciences, 15(11), 527–536. [Link]

Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where Is the Semantic

System? A Critical Review and Meta-Analysis of 120 Functional Neuroimaging

Studies. Cerebral Cortex, 19(12), 2767–2796. [Link]

Binney, R. J., Embleton, K. V., Jefferies, E., Parker, G. J. M., & Lambon Ralph, M. A. (2010).

The Ventral and Inferolateral Aspects of the Anterior Temporal Lobe Are Crucial in

Semantic Memory: Evidence from a Novel Direct Comparison of Distortion-

Corrected fMRI, rTMS, and Semantic Dementia. Cerebral Cortex, 20(11), 2728–

2738. [Link]

Binney, R. J., Parker, G. J. M., & Lambon Ralph, M. A. (2012). Convergent Connectivity and

Graded Specialization in the Rostral Human Temporal Lobe as Revealed by

Diffusion-Weighted Imaging Probabilistic Tractography. Journal of Cognitive

Neuroscience, 24(10), 1998–2014. [Link]

Bopp, K. L., & Verhaeghen, P. (2018). Aging and n-Back Performance: A Meta-Analysis. The

Journals of Gerontology: Series B. [Link]

Bourguignon, N. J., & Gracco, V. L. (2019). A dual architecture for the cognitive control of

language: Evidence from functional imaging and language production. NeuroImage,

192, 26–37. [Link]

Bourguignon, N. J., Ohashi, H., Nguyen, D., & Gracco, V. L. (2018). The neural dynamics of

competition resolution for language production in the prefrontal cortex. Human Brain

Mapping, 39(3), 1391–1402. [Link]

178
Bower, J., Acar, S., & Kursuncu, U. (2023). Measuring Creativity in Academic Writing: An

Analysis of Essays in Advanced Placement Language and Composition. Journal of

Advanced Academics, 34(3–4), 183–214.

[Link]

Bozeat, S., Lambon Ralph, M. A., Patterson, K., Garrard, P., & Hodges, J. R. (2000). Non-

verbal semantic impairment in semantic dementia. Neuropsychologia, 38(9), 1207–

1215. [Link]

Bozeat, S., Ralph, M. A. L., Patterson, K., & Hodges, J. R. (2002). When objects lose their

meaning: What happens to their use? Cognitive, Affective, & Behavioral

Neuroscience, 2(3), 236–251. [Link]

Branzi, F. M., & Lambon Ralph, M. A. (2023). Semantic‐specific and domain‐general

mechanisms for integration and update of contextual information. Human Brain

Mapping, 44(17), 5547–5566. [Link]

Brass, M., & von Cramon, D. Y. (2004). Selection for cognitive control: A functional

magnetic resonance imaging study on the selection of task-relevant information.

Journal of Neuroscience, 24(40), 8847–8852.

Brown, T. B. (2020). Language models are few-shot learners. arXiv Preprint

arXiv:2005.14165.

Buckner, R. L. (2012). The serendipitous discovery of the brain’s default network.

NeuroImage, 62(2), 1137–1145. [Link]

Buczak, P., Huang, H., Forthmann, B., & Doebler, P. (2023). The Machines Take Over: A

Comparison of Various Supervised Learning Approaches for Automated Scoring of

Divergent Thinking Tasks. The Journal of Creative Behavior, 57(1), 17–36.

[Link]

179
Calhoun, V. D., Liu, J., & Adalı, T. (2009). A review of group ICA for fMRI data and ICA for

joint inference of imaging, genetic, and ERP data. NeuroImage, 45(1), S163–S172.

[Link]

Campanella, F., Mondani, M., Skrap, M., & Shallice, T. (2009). Semantic access dysphasia

resulting from left temporal lobe tumours. Brain, 132(1), 87–102.

[Link]

Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-analytic studies.

Cambridge university press.

Cattell, R. B. (1971). Abilities: Their structure, growth, and action.

Chen, Y., Chen, K., Ding, J., Zhang, Y., Yang, Q., Lv, Y., Guo, Q., & Han, Z. (2019). Neural

substrates of amodal and modality-specific semantic processing within the temporal

lobe: A lesion-behavior mapping study of semantic dementia. Cortex, 120, 78–91.

[Link]

Cheng, L., Hu, W., Jia, X., & Runco, M. A. (2016). The different role of cognitive inhibition

in early versus late creative problem finding. Psychology of Aesthetics, Creativity, and

the Arts, 10(1), 32–41. [Link]

Chiou, R., Humphreys, G. F., Jung, J., & Lambon Ralph, M. A. (2018). Controlled semantic

cognition relies upon dynamic and flexible interactions between the executive

‘semantic control’ and hub-and-spoke ‘semantic representation’ systems. Cortex, 103,

100–116. [Link]

Chiou, R., Jefferies, E., Duncan, J., Humphreys, G. F., & Lambon Ralph, M. A. (2023a). A

middle ground where executive control meets semantics: The neural substrates of

semantic control are topographically sandwiched between the multiple-demand and

default-mode systems. Cerebral Cortex, 33(8), 4512–4526.

[Link]

180
Chiou, R., Jefferies, E., Duncan, J., Humphreys, G. F., & Lambon Ralph, M. A. (2023b). A

middle ground where executive control meets semantics: The neural substrates of

semantic control are topographically sandwiched between the multiple-demand and

default-mode systems. Cerebral Cortex, 33(8), 4512–4526.

[Link]

Chiou, R., & Lambon Ralph, M. A. (2019). Unveiling the dynamic interplay between the

hub- and spoke-components of the brain’s semantic system and its impact on human

behaviour. NeuroImage, 199, 114–126.

[Link]

Cho, S. H., Nijenhuis, J. T., Vianen, A. E. M., Kim, H.-B., & Lee, K. H. (2010). The

Relationship Between Diverse Components of Intelligence and Creativity. The

Journal of Creative Behavior, 44(2), 125–137. [Link]

6057.2010.tb01329.x

Christoff, K., Gordon, A. M., Smallwood, J., Smith, R., & Schooler, J. W. (2009). Experience

sampling during fMRI reveals default network and executive system contributions to

mind wandering. Proceedings of the National Academy of Sciences, 106(21), 8719–

8724. [Link]

Christoff, K., Irving, Z. C., Fox, K. C. R., Spreng, R. N., & Andrews-Hanna, J. R. (2016).

Mind-wandering as spontaneous thought: A dynamic framework. Nature Reviews

Neuroscience, 17(11), 718–731. [Link]

Chrysikou, E. G. (2019). Creativity in and out of (cognitive) control. Current Opinion in

Behavioral Sciences, 27, 94–99. [Link]

Coccia, M., Bartolini, M., Luzzi, S., Provinciali, L., & Lambon Ralph, M. A. (2004).

Semantic memory is an amodal, dynamic system: Evidence from the interaction of

181
 naming and object use in semantic dementia. Cognitive Neuropsychology, 21(5), 513–

527. [Link]

Cogdell‐Brooke, L. S., Sowden, P. T., Violante, I. R., & Thompson, H. E. (2020). A meta-

analysis of functional magnetic resonance imaging studies of divergent thinking using

activation likelihood estimation. Human Brain Mapping, 41(17), 5057–5077.

[Link]

Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing.

Psychological Review, 82, 407–428. [Link]

Cona, G., Santacesaria, P., & Scarpazza, C. (2023). Envisioning the future: An ALE meta-

analysis on neural correlates of future thinking, prospective memory and delay

discounting. Neuroscience & Biobehavioral Reviews, 153, 105355.

[Link]

Corbett, F., Jefferies, E., & Lambon Ralph, M. A. (2009). Exploring multimodal semantic

control impairments in semantic aphasia: Evidence from naturalistic object use.

Neuropsychologia, 47(13), 2721–2731.

[Link]

Cortes, R. A., Weinberger, A. B., Daker, R. J., & Green, A. E. (2019). Re-examining

prominent measures of divergent and convergent creativity. Current Opinion in

Behavioral Sciences, 27, 90–93. [Link]

Cosgrove, A. L., Kenett, Y. N., Beaty, R. E., & Diaz, M. T. (2021). Quantifying flexibility in

thought: The resiliency of semantic networks differs across the lifespan. Cognition,

211, 104631. [Link]

Coutanche, M. N., & Thompson-Schill, S. L. (2015). Creating Concepts from Converging

Features in Human Cortex. Cerebral Cortex, 25(9), 2584–2593.

[Link]

182
Cusack, R., Cumming, N., Bor, D., Norris, D., & Lyzenga, J. (2005). Automated post-hoc

noise cancellation tool for audio recordings acquired in an MRI scanner. Human

Brain Mapping, 24(4), 299–304.

Damasio, A. R. (1989). The brain binds entities and events by multiregional activation from

convergence zones. Neural Computation, 1(1), 123–132.

Davelaar, E. J. (2015). Semantic Search in the Remote Associates Test. Topics in Cognitive

Science, 7(3), 494–512. [Link]

Davey, J., Cornelissen, P. L., Thompson, H. E., Sonkusare, S., Hallam, G., Smallwood, J., &

Jefferies, E. (2015). Automatic and Controlled Semantic Retrieval: TMS Reveals

Distinct Contributions of Posterior Middle Temporal Gyrus and Angular Gyrus. The

Journal of Neuroscience, 35(46), 15230–15239.

[Link]

Davey, J., Thompson, H. E., Hallam, G., Karapanagiotidis, T., Murphy, C., De Caso, I.,

Krieger-Redwood, K., Bernhardt, B. C., Smallwood, J., & Jefferies, E. (2016).

Exploring the role of the posterior middle temporal gyrus in semantic cognition:

Integration of anterior temporal lobe with executive processes. NeuroImage, 137,

165–177. [Link]

De Dreu, C. K. W., Nijstad, B. A., Baas, M., Wolsink, I., & Roskes, M. (2012). Working

Memory Benefits Creative Insight, Musical Improvisation, and Original Ideation

Through Maintained Task-Focused Attention. Personality and Social Psychology

Bulletin, 38(5), 656–669. [Link]

Delis, D. C., Kaplan, E., & Kramer, J. H. (2001). Delis-Kaplan Executive Function System

(D-KEFS). The Psychological Corporation.

183
Devlin, J. T., Russell, R. P., Davis, M. H., Price, C. J., Wilson, J., Moss, H. E., Matthews, P.

M., & Tyler, L. K. (2000). Susceptibility-induced loss of signal: Comparing PET and

fMRI on a semantic task. Neuroimage, 11(6), 589–600.

Diamond, A. (2013). Executive Functions. Annual Review of Psychology, 64(1), 135–168.

[Link]

Diedrich, J., Jauk, E., Silvia, P. J., Gredlein, J. M., Neubauer, A. C., & Benedek, M. (2018).

Assessment of real-life creativity: The Inventory of Creative Activities and

Achievements (ICAA). Psychology of Aesthetics, Creativity, and the Arts, 12(3), 304–

316. [Link]

Diveica, V., Koldewyn, K., & Binney, R. J. (2021). Establishing a role of the semantic control

network in social cognitive processing: A meta-analysis of functional neuroimaging

studies. NeuroImage, 245, 118702. [Link]

Dobbins, I. G., & Wagner, A. D. (2005). Domain-general and domain-sensitive prefrontal

mechanisms for recollecting events and detecting novelty. Cerebral Cortex, 15(11),

1768–1778.

Dollinger, S. J. (2003). Need for Uniqueness, Need for Cognition, and Creativity. The

Journal of Creative Behavior, 37(2), 99–116. [Link]

6057.2003.tb00828.x

D’Souza, R. (2021a). What characterises creativity in narrative writing, and how do we

assess it? Research findings from a systematic literature search. Thinking Skills and

Creativity, 42, 100949. [Link]

D’Souza, R. (2021b). What characterises creativity in narrative writing, and how do we

assess it? Research findings from a systematic literature search. Thinking Skills and

Creativity, 42, 100949. [Link]

184
Dumas, D., Organisciak, P., & Doherty, M. (2021). Measuring divergent thinking originality

with human raters and text-mining models: A psychometric comparison of methods.

Psychology of Aesthetics, Creativity, and the Arts, 15(4), 645–663.

[Link]

Dumas, D., Organisciak, P., Maio, S., & Doherty, M. (2021). Four Text‐Mining Methods for

Measuring Elaboration. The Journal of Creative Behavior, 55(2), 517–531.

[Link]

Duncan, J. (2010). The multiple-demand (MD) system of the primate brain: Mental programs

for intelligent behaviour. Trends in Cognitive Sciences, 14(4), 172–179.

[Link]

DuPre, E., Salo, T., Ahmed, Z., Bandettini, P. A., Bottenhorn, K. L., Caballero-Gaudes, C.,

Dowdle, L. T., Gonzalez-Castillo, J., Heunis, S., Kundu, P., & others. (2021). TE-

dependent analysis of multi-echo fMRI with* tedana. Journal of Open Source

Software, 6(66), 3669.

Dygert, S. K. C., & Jarosz, A. F. (2020). Individual differences in creative cognition. Journal

of Experimental Psychology: General, 149(7), 1249–1274.

[Link]

Eggert, G. H., & Wernicke, C. (1977). Wernicke’s works on aphasia: A sourcebook and

review. Mouton; WorldCat.

Ellamil, M., Dobson, C., Beeman, M., & Christoff, K. (2012). Evaluative and generative

modes of thought during the creative process. NeuroImage, 59(2), 1783–1794.

[Link]

Ellis, D. M., & Brewer, G. A. (2018). Aiding the search: Examining individual differences in

multiply-constrained problem solving. Consciousness and Cognition, 62, 21–33.

[Link]

185
Ellis, D. M., & Robison, M. (n.d.). The Cognitive Underpinnings of Multiply-Constrained

Problem Solving.

Ellis, D. M., Robison, M. K., & Brewer, G. A. (2021). The Cognitive Underpinnings of

Multiply-Constrained Problem Solving. Journal of Intelligence, 9(1).

[Link]

Evans, J. S. B. (2008). Dual-processing accounts of reasoning, judgment, and social

cognition. Annu. Rev. Psychol., 59(1), 255–278.

Eysenck, H. J. (2003). Creativity, personality and the convergent-divergent continuum. In

Critical creative processes. (pp. 95–114). Hampton Press.

Fan, L., Zhuang, K., Wang, X., Zhang, J., Liu, C., Gu, J., & Qiu, J. (2023). Exploring the

behavioral and neural correlates of semantic distance in creative writing.

Psychophysiology, 60(5), e14239. [Link]

Farahibozorg, S.-R., Henson, R. N., Woollams, A. M., & Hauk, O. (2022). Distinct roles for

the anterior temporal lobe and angular gyrus in the spatiotemporal cortical semantic

network. Cerebral Cortex, 32(20), 4549–4564.

[Link]

Faust, M., & Kenett, Y. N. (2014). Rigidity, chaos and integration: Hemispheric interaction

and individual differences in metaphor comprehension. Frontiers in Human

Neuroscience, 8. [Link]

Fedorenko, E., & Blank, I. A. (2020). Broca’s Area Is Not a Natural Kind. Trends in

Cognitive Sciences, 24(4), 270–284. [Link]

Fedorenko, E., Duncan, J., & Kanwisher, N. (2013). Broad domain generality in focal regions

of frontal and parietal cortex. Proceedings of the National Academy of Sciences,

110(41), 16616–16621. [Link]

186
Feinberg, D. A., Moeller, S., Smith, S. M., Auerbach, E., Ramanna, S., Glasser, M. F., Miller,

K. L., Ugurbil, K., & Yacoub, E. (2010). Multiplexed echo planar imaging for sub-

second whole brain FMRI and fast diffusion imaging. PloS One, 5(12), e15710.

Fink, A., Benedek, M., Koschutnig, K., Papousek, I., Weiss, E. M., Bagga, D., & Schöpf, V.

(2018). Modulation of resting-state network connectivity by verbal divergent thinking

training. Brain and Cognition, 128, 1–6. [Link]

Finke, R. A., Ward, T. B., & Smith, S. M. (1992). Creative cognition: Theory, research, and

applications. (pp. vi, 239). The MIT Press.

Foos, P. W., & Boone, D. (2008). Adult Age Differences in Divergent Thinking: It’s Just a

Matter of Time. Educational Gerontology, 34(7), 587–594.

[Link]

Forthmann, B., Gerwig, A., Holling, H., Çelik, P., Storme, M., & Lubart, T. (2016). The be-

creative effect in divergent thinking: The interplay of instruction and object frequency.

Intelligence, 57, 25–32. [Link]

Forthmann, B., Jendryczko, D., Scharfen, J., Kleinkorres, R., Benedek, M., & Holling, H.

(2019). Creative ideation, broad retrieval ability, and processing speed: A

confirmatory study of nested cognitive abilities. Intelligence, 75, 59–72.

[Link]

Forthmann, B., Oyebade, O., Ojo, A., Günther, F., & Holling, H. (2019). Application of

Latent Semantic Analysis to Divergent Thinking is Biased by Elaboration. The

Journal of Creative Behavior, 53(4), 559–575. [Link]

Forthmann, B., Paek, S. H., Dumas, D., Barbot, B., & Holling, H. (2020). Scrutinizing the

basis of originality in divergent thinking tests: On the measurement precision of

response propensity estimates. British Journal of Educational Psychology, 90(3),

683–699. [Link]

187
Forthmann, B., Szardenings, C., & Holling, H. (2020). Understanding the confounding effect

of fluency in divergent thinking scores: Revisiting average scores to quantify

artifactual correlation. Psychology of Aesthetics, Creativity, and the Arts, 14(1), 94–

112. [Link]

Fox, K. C. R., Spreng, R. N., Ellamil, M., Andrews-Hanna, J. R., & Christoff, K. (2015). The

wandering brain: Meta-analysis of functional neuroimaging studies of mind-

wandering and related spontaneous thought processes. NeuroImage, 111, 611–621.

[Link]

Fox, M. D., Snyder, A. Z., Vincent, J. L., Corbetta, M., Van Essen, D. C., & Raichle, M. E.

(2005). The human brain is intrinsically organized into dynamic, anticorrelated

functional networks. Proceedings of the National Academy of Sciences, 102(27),

9673–9678. [Link]

Fradkin, I., & Eldar, E. (2023). Accumulating evidence for myriad alternatives: Modeling the

generation of free association. Psychological Review, 130(6), 1492–1520.

[Link]

Friedman, N. P., Miyake, A., Altamirano, L. J., Corley, R. P., Young, S. E., Rhea, S. A., &

Hewitt, J. K. (2016). Stability and change in executive function abilities from late

adolescence to early adulthood: A longitudinal twin study. Developmental

Psychology, 52(2), 326–340. [Link]

Friedman, N. P., Miyake, A., Young, S. E., DeFries, J. C., Corley, R. P., & Hewitt, J. K.

(2008). Individual differences in executive functions are almost entirely genetic in

origin. Journal of Experimental Psychology: General, 137(2), 201.

Frith, E., Elbich, D. B., Christensen, A. P., Rosenberg, M. D., Chen, Q., Kane, M. J., Silvia, P.

J., Seli, P., & Beaty, R. E. (2021). Intelligence and Creativity Share a Common

Cognitive and Neural Basis.

188
Frith, E., Miller, S., & Loprinzi, P. D. (2020). A Review of Experimental Research on

Embodied Creativity: Revisiting the Mind–Body Connection. The Journal of Creative

Behavior, 54(4), 767–798. [Link]

Frost, A., Moussaoui, S., Kaur, J., Aziz, S., Fukuda, K., & Niemeier, M. (2021). Is the n-back

task a measure of unstructured working memory capacity? Towards understanding its

connection to other working memory tasks. Acta Psychologica, 219, 103398.

[Link]

Fürst, G., & Grin, F. (2018). A comprehensive method for the measurement of everyday

creativity. Thinking Skills and Creativity, 28, 84–97.

[Link]

Fusi, G., Lavolpe, S., Crepaldi, M., & Rusconi, M. L. (2021). The Controversial Effect of

Age on Divergent Thinking Abilities: A Systematic Review. The Journal of Creative

Behavior, 55(2), 374–395. [Link]

Gabora, L. (2010). Honing Theory: A Complex Systems Framework for Creativity.

Gabora, L. (2018). The Neural Basis and Evolution of Divergent and Convergent Thought. In

R. E. Jung & O. Vartanian (Eds.), The Cambridge Handbook of the Neuroscience of

Creativity (1st ed., pp. 58–70). Cambridge University Press.

[Link]

Gajda, A., Karwowski, M., & Beghetto, R. A. (2017). Creativity and academic achievement:

A meta-analysis. Journal of Educational Psychology, 109(2), 269.

Galton, C. J. (2000). Atypical and typical presentations of Alzheimer’s disease: A clinical,

neuropsychological, neuroimaging and pathological study of 13 cases. Brain, 123(3),

484–498. [Link]

Gao, Z., Zheng, L., Chiou, R., Gouws, A., Krieger-Redwood, K., Wang, X., Varga, D., Ralph,

M. A. L., Smallwood, J., & Jefferies, E. (2021). Distinct and common neural coding

189
 of semantic and non-semantic control demands. NeuroImage, 236, 118230.

[Link]

Gardner, H. E., Lambon Ralph, M. A., Dodds, N., Jones, T., Ehsan, S., & Jefferies, E. (2012).

The Differential Contributions of pFC and Temporo-parietal Cortex to Multimodal

Semantic Control: Exploring Refractory Effects in Semantic Aphasia. Journal of

Cognitive Neuroscience, 24(4), 778–793. [Link]

Geranmayeh, F., Leech, R., & Wise, R. J. S. (2016). Network dysfunction predicts speech

production after left hemisphere stroke. Neurology, 86(14), 1296–1305.

[Link]

Geranmayeh, F., Wise, R. J. S., Mehta, A., & Leech, R. (2014). Overlapping Networks

Engaged during Spoken Language Production and Its Cognitive Control. Journal of

Neuroscience, 34(26), 8728–8740. [Link]

14.2014

Gerlach, K. D., Spreng, R. N., Madore, K. P., & Schacter, D. L. (2014). Future planning:

Default network activity couples with frontoparietal control network and reward-

processing regions during process and outcome simulations. Social Cognitive and

Affective Neuroscience, 9(12), 1942–1951. [Link]

Gerver, C. R., Griffin, J. W., Dennis, N. A., & Beaty, R. E. (2023). Memory and creativity: A

meta-analytic examination of the relationship between memory systems and creative

cognition. Psychonomic Bulletin & Review, 30(6), 2116–2154.

[Link]

Gerwig, A., Miroshnik, K., Forthmann, B., Benedek, M., Karwowski, M., & Holling, H.

(2021). The Relationship between Intelligence and Divergent Thinking—A Meta-

Analytic Update. Journal of Intelligence, 9(2), 23.

[Link]

190
Giffard, B. (2001). The nature of semantic memory deficits in Alzheimer’s disease: New

insights from hyperpriming effects. Brain, 124(8), 1522–1532.

[Link]

Gilhooly, K. J., Fioratou, E., Anthony, S. H., & Wynn, V. (2007). Divergent thinking:

Strategies and executive involvement in generating novel uses for familiar objects.

British Journal of Psychology, 98(4), 611–625. [Link]

8295.2007.tb00467.x

Glosser, G., & Deser, T. (1992). A Comparison of Changes in Macrolinguistic and

Microlinguistic Aspects of Discourse Production in Normal Aging. Journal of

Gerontology, 47(4), P266–P272. [Link]

Gonen-Yaacovi, G., De Souza, L. C., Levy, R., Urbanski, M., Josse, G., & Volle, E. (2013).

Rostral and caudal prefrontal contribution to creativity: A meta-analysis of functional

imaging data. Frontiers in Human Neuroscience, 7.

[Link]

Gonzalez Alam, T. R. D. J., Karapanagiotidis, T., Smallwood, J., & Jefferies, E. (2019).

Degrees of lateralisation in semantic cognition: Evidence from intrinsic connectivity.

NeuroImage, 202, 116089. [Link]

Goulden, N., Khusnulina, A., Davis, N. J., Bracewell, R. M., Bokde, A. L., McNulty, J. P., &

Mullins, P. G. (2014). The salience network is responsible for switching between the

default mode network and the central executive network: Replication from DCM.

NeuroImage, 99, 180–190. [Link]

Gray, K., Anderson, S., Chen, E. E., Kelly, J. M., Christian, M. S., Patrick, J., Huang, L.,

Kenett, Y. N., & Lewis, K. (2019). “Forward flow”: A new measure to quantify free

thought and predict creativity. American Psychologist, 74(5), 539.

191
Green, A. E., Beaty, R. E., Kenett, Y. N., & Kaufman, J. C. (2023). The Process Definition of

Creativity. Creativity Research Journal, 1–29.

[Link]

Green, A. E., Cohen, M. S., Raab, H. A., Yedibalian, C. G., & Gray, J. R. (2015). Frontopolar

activity and connectivity support dynamic conscious augmentation of creative state.

Human Brain Mapping, 36(3), 923–934. [Link]

Grohman, M., Wodniecka, Z., & Kłusak, M. (2006). Divergent Thinking and Evaluation

Skills: Do They Always Go Together? The Journal of Creative Behavior, 40(2), 125–

145. [Link]

Grossman, M. (2003). Neural basis for semantic memory difficulty in Alzheimer’s disease:

An fMRI study. Brain, 126(2), 292–311. [Link]

Gruenewald, P. J., & Lockhead, G. R. (1980). The free recall of category examples. Journal

of Experimental Psychology: Human Learning and Memory, 6(3), 225.

Gruszka, A., & Necka, E. (2002). Priming and Acceptance of Close and Remote Associations

by Creative and Less Creative People. Creativity Research Journal, 14(2), 193–205.

[Link]

Guilford, J. P. (1950). Creativity. American Psychologist, 5(9), 444–454.

[Link]

Guilford, J. P. (1956). The structure of intellect. Psychological Bulletin, 53(4), 267.

Guilford, J. P. (1967). Creativity: Yesterday, Today and Tomorrow. The Journal of Creative

Behavior, 1(1), 3–14. [Link]

Guilford, J. P., Christensen, P. R., Merrifield, P. R., & Wilson, R. C. (1978). Alternate uses.

Guo, C. C., Gorno-Tempini, M. L., Gesierich, B., Henry, M., Trujillo, A., Shany-Ur, T.,

Jovicich, J., Robinson, S. D., Kramer, J. H., Rankin, K. P., Miller, B. L., & Seeley, W.

192
 W. (2013). Anterior temporal lobe degeneration produces widespread network-driven

dysfunction. Brain, 136(10), 2979–2991. [Link]

Guo, J., Ge, Y., & Pang, W. (2019). The underlying cognitive mechanisms of the rater effect

in creativity assessment: The mediating role of perceived semantic distance. Thinking

Skills and Creativity, 33, 100572. [Link]

Guo, Y., Lin, S., Acar, S., Jin, S., Xu, X., Feng, Y., & Zeng, Y. (2022). Divergent Thinking

and Evaluative Skill: A Meta‐Analysis. The Journal of Creative Behavior, 56(3), 432–

448. [Link]

Gupta, N., Jang, Y., Mednick, S. C., & Huber, D. E. (2012). The Road Not Taken: Creative

Solutions Require Avoidance of High-Frequency Responses. Psychological Science,

23(3), 288–294. [Link]

Hallam, G. P., Whitney, C., Hymers, M., Gouws, A. D., & Jefferies, E. (2016). Charting the

effects of TMS with fMRI: Modulation of cortical recruitment within the distributed

network supporting semantic control. Neuropsychologia, 93, 40–52.

[Link]

Hass, R. W. (2017). Semantic search during divergent thinking. Cognition, 166, 344–357.

[Link]

Hassabis, D., & Maguire, E. A. (2007). Deconstructing episodic memory with construction.

Trends in Cognitive Sciences, 11(7), 299–306.

[Link]

He, L., Kenett, Y. N., Zhuang, K., Liu, C., Zeng, R., Yan, T., Huo, T., & Qiu, J. (2021). The

relation between semantic memory structure, associative abilities, and verbal and

figural creativity. Thinking & Reasoning, 27(2), 268–293.

[Link]

193
He, R., Zhuang, K., Liu, L., Ding, K., Wang, X., Fu, L., Qiu, J., & Chen, Q. (2022). The

impact of knowledge on poetry composition: An fMRI investigation. Brain and

Language, 235, 105202. [Link]

Head, H. (1926). Aphasia and kindred disorders of speech (Vol. 1). CUP Archive.

Hennessey, B. A., & Amabile, T. M. (2010). Creativity. Annual Review of Psychology, 61(1),

569–598. [Link]

Henry, J. D., & Crawford, J. R. (2004). A Meta-Analytic Review of Verbal Fluency

Performance in Patients With Traumatic Brain Injury. Neuropsychology, 18(4), 621–

628. [Link]

Hills, T. T., & Kenett, Y. N. (2022). Is the Mind a Network? Maps, Vehicles, and Skyhooks in

Cognitive Network Science. Topics in Cognitive Science, 14(1), 189–208.

[Link]

Hills, T. T., Todd, P. M., & Goldstone, R. L. (2008). Search in External and Internal Spaces:

Evidence for Generalized Cognitive Search Processes. Psychological Science, 19(8),

802–808. [Link]

Hills, T. T., Todd, P. M., & Goldstone, R. L. (2010). The central executive as a search

process: Priming exploration and exploitation across domains. Journal of

Experimental Psychology: General, 139(4), 590–609.

[Link]

Hills, T. T., Todd, P. M., & Jones, M. N. (2012). Optimal Foraging in Semantic Memory.

Hills, T. T., Todd, P. M., & Jones, M. N. (2015). Foraging in Semantic Fields: How We

Search Through Memory. Topics in Cognitive Science, 7(3), 513–534.

[Link]

194
Hills, T. T., Todd, P. M., Lazer, D., Redish, A. D., & Couzin, I. D. (2015). Exploration versus

exploitation in space, mind, and society. Trends in Cognitive Sciences, 19(1), 46–54.

[Link]

Hocevar, D. (1979). The Development dfpe Creative Behavior Inventory (CBI).

Hodges, J. R., Graham, N., & Patterson, K. (1995). Charting the progression in semantic

dementia: Implications for the organisation of semantic memory. Memory, 3(3–4),

463–495. [Link]

Hodges, J. R., & Patterson, K. (2007). Semantic dementia: A unique clinicopathological

syndrome. The Lancet Neurology, 6(11), 1004–1014. [Link]

4422(07)70266-1

Hodges, J. R., Patterson, K., Oxbury, S., & Funnell, E. (1992). Semantic Dementia:

Progressive fluent apahsia with temporal lobe atrophy. Brain, 115(6), 1783–1806.

[Link]

Hodgson, V. J., Lambon Ralph, M. A., & Jackson, R. L. (2021). Multiple dimensions

underlying the functional organization of the language network. NeuroImage, 241,

118444. [Link]

Hoenig, K., Sim, E.-J., Bochev, V., Herrnberger, B., & Kiefer, M. (2008). Conceptual

Flexibility in the Human Brain: Dynamic Recruitment of Semantic Maps from Visual,

Motor, and Motion-related Areas. Journal of Cognitive Neuroscience, 20(10), 1799–

1814. [Link]

Hoffman, P. (2018). An individual differences approach to semantic cognition: Divergent

effects of age on representation, retrieval and selection. Scientific Reports, 8(1), 8145.

[Link]

195
Hoffman, P. (2019). Reductions in prefrontal activation predict off-topic utterances during

speech production. Nature Communications, 10(1), 515.

[Link]

Hoffman, P., Cogdell-Brooke, L., & Thompson, H. E. (2020). Going off the rails: Impaired

coherence in the speech of patients with semantic control deficits. Neuropsychologia,

146, 107516. [Link]

Hoffman, P., Jefferies, E., & Lambon Ralph, M. A. (2011). Remembering ‘zeal’ but not

‘thing’: Reverse frequency effects as a consequence of deregulated semantic

processing. Neuropsychologia, 49(3), 580–584.

[Link]

Hoffman, P., Jones, R. W., & Ralph, M. A. L. (2012). The degraded concept representation

system in semantic dementia: Damage to pan-modal hub, then visual spoke. Brain,

135(12), 3770–3780. [Link]

Hoffman, P., & Lambon Ralph, M. A. (2011). Reverse Concreteness Effects Are Not a

Typical Feature of Semantic Dementia: Evidence for the Hub-and-Spoke Model of

Conceptual Representation. Cerebral Cortex, 21(9), 2103–2112.

[Link]

Hoffman, P., Loginova, E., & Russell, A. (2018). Poor coherence in older people’s speech is

explained by impaired semantic and executive processes.

Hoffman, P., & MacPherson, S. E. (2022). What determines cognitive estimation ability?

Changing contributions of semantic and executive domains as a function of age.

Journal of Neuropsychology, 16(3), 481–497. [Link]

Hoffman, P., McClelland, J. L., & Lambon Ralph, M. A. (2018). Concepts, control, and

context: A connectionist account of normal and disordered semantic cognition.

Psychological Review, 125(3), 293–328. [Link]

196
Hoffman, P., Rogers, T. T., & Lambon Ralph, M. A. (2011). Semantic Diversity Accounts for

the “Missing” Word Frequency Effect in Stroke Aphasia: Insights Using a Novel

Method to Quantify Contextual Variability in Meaning. Journal of Cognitive

Neuroscience, 23(9), 2432–2446. [Link]

Homack, S., Lee, D., & Riccio, C. A. (2005). Test Review: Delis-Kaplan Executive Function

System. Journal of Clinical and Experimental Neuropsychology, 27(5), 599–609.

[Link]

Horn, J. L. (1965). A rationale and test for the number of factors in factor analysis.

Psychometrika, 30(2), 179–185. [Link]

Howard, Z. L., Belevski, B., Eidels, A., & Dennis, S. (2020). What do cows drink? A systems

factorial technology account of processing architecture in memory intersection

problems. Cognition, 202, 104294. [Link]

Howard-Jones, P. A., Blakemore, S.-J., Samuel, E. A., Summers, I. R., & Claxton, G. (2005).

Semantic divergence and creative story generation: An fMRI investigation. Cognitive

Brain Research, 25(1), 240–250. [Link]

Hoyer, W. J., Stawski, R. S., Wasylyshyn, C., & Verhaeghen, P. (2004). Adult Age and Digit

Symbol Substitution Performance: A Meta-Analysis. Psychology and Aging, 19(1),

211–214. [Link]

Humphreys, G. F., Hoffman, P., Visser, M., Binney, R. J., & Lambon Ralph, M. A. (2015).

Establishing task- and modality-dependent dissociations between the semantic and

default mode networks. Proceedings of the National Academy of Sciences, 112(25),

7857–7862. [Link]

Humphreys, G. F., & Lambon Ralph, M. A. (2017). Mapping Domain-Selective and

Counterpointed Domain-General Higher Cognitive Functions in the Lateral Parietal

Cortex: Evidence from fMRI Comparisons of Difficulty-Varying Semantic Versus

197
 Visuo-Spatial Tasks, and Functional Connectivity Analyses. Cerebral Cortex, 27(8),

4199–4212. [Link]

Humphreys, G. F., Lambon Ralph, M. A., & Simons, J. S. (2021). A Unifying Account of

Angular Gyrus Contributions to Episodic and Semantic Cognition. Trends in

Neurosciences, 44(6), 452–463. [Link]

Idowu, M. I., & Szameitat, A. J. (2023). Executive function abilities in cognitively healthy

young and older adults—A cross-sectional study. Frontiers in Aging Neuroscience,

15, 976915. [Link]

Ikeda, M., Patterson, K., Graham, K. S., Ralph, M. A. L., & Hodges, J. R. (2006). A horse of

a different colour: Do patients with semantic dementia recognise different versions of

the same object as the same? Neuropsychologia, 44(4), 566–575.

[Link]

Ilha Villanova, A. L., & Pina E Cunha, M. (2021). Everyday Creativity: A Systematic

Literature Review. The Journal of Creative Behavior, 55(3), 673–695.

[Link]

Irish, M., & Vatansever, D. (2020). Rethinking the episodic-semantic distinction from a

gradient perspective. Current Opinion in Behavioral Sciences, 32, 43–49.

[Link]

Ivancovsky, T., Shamay-Tsoory, S., Lee, J., Morio, H., & Kurman, J. (2019). A dual process

model of generation and evaluation: A theoretical framework to examine cross-

cultural differences in the creative process. Personality and Individual Differences,

139, 60–68. [Link]

Jackson, R. L. (2021). The neural correlates of semantic control revisited. NeuroImage, 224,

117444. [Link]

198
Jackson, R. L., Hoffman, P., Pobric, G., & Lambon Ralph, M. A. (2015). The Nature and

Neural Correlates of Semantic Association versus Conceptual Similarity. Cerebral

Cortex, 25(11), 4319–4333. [Link]

James, L. E., Burke, D. M., Austin, A., & Hulme, E. (1998). Production and perception of"

verbosity" in younger and older adults. Psychology and Aging, 13(3), 355.

Jefferies, E., Baker, S. S., Doran, M., & Ralph, M. A. L. (2007). Refractory effects in stroke

aphasia: A consequence of poor semantic control. Neuropsychologia, 45(5), 1065–

1079. [Link]

Jefferies, E., & Lambon Ralph, M. A. (2006). Semantic impairment in stroke aphasia versus

semantic dementia: A case-series comparison. Brain, 129(8), 2132–2147.

[Link]

Jefferies, E., Patterson, K., Jones, R. W., & Lambon Ralph, M. A. (2009a). Comprehension of

concrete and abstract words in semantic dementia. Neuropsychology, 23(4), 492–499.

[Link]

Jefferies, E., Patterson, K., Jones, R. W., & Lambon Ralph, M. A. (2009b). Comprehension of

concrete and abstract words in semantic dementia. Neuropsychology, 23(4), 492–499.

[Link]

Jefferies, E., Patterson, K., & Ralph, M. A. L. (2008). Deficits of knowledge versus executive

control in semantic cognition: Insights from cued naming. Neuropsychologia, 46(2),

649–658. [Link]

Jefferies, E., Thompson, H., Cornelissen, P., & Smallwood, J. (2020). The neurocognitive

basis of knowledge about object identity and events: Dissociations reflect opposing

effects of semantic coherence and control. Philosophical Transactions of the Royal

Society B: Biological Sciences, 375(1791), 20190300.

[Link]

199
Jiang, Y., & Kanwisher, N. (2003). Common Neural Substrates for Response Selection across

Modalities and Mapping Paradigms. Journal of Cognitive Neuroscience, 15(8), 1080–

1094. [Link]

Johnson, D. R., Kaufman, J. C., Baker, B. S., Patterson, J. D., Barbot, B., Green, A. E., Van

Hell, J., Kennedy, E., Sullivan, G. F., Taylor, C. L., Ward, T., & Beaty, R. E. (2022).

Divergent semantic integration (DSI): Extracting creativity from narratives with

distributional semantic modeling. Behavior Research Methods, 55(7), 3726–3759.

[Link]

Jonides, J., & Nee, D. E. (2006). Brain mechanisms of proactive interference in working

memory. Neuroscience, 139(1), 181–193.

Jung, J., Lambon Ralph, M. A., & Jackson, R. L. (2022). Subregions of DLPFC Display

Graded yet Distinct Structural and Functional Connectivity. The Journal of

Neuroscience, 42(15), 3241–3252. [Link]

21.2022

Kahneman, D. (2011). Thinking, fast and slow. Macmillan.

Kaufman, J. C. (2012). Counting the muses: Development of the Kaufman Domains of

Creativity Scale (K-DOCS). Psychology of Aesthetics, Creativity, and the Arts, 6(4),

298–308. [Link]

Kaufman, J. C. (2016). Creativity 101. Springer publishing company.

Kaufman, J. C., & Plucker, J. A. (2011). Intelligence and Creativity. In R. J. Sternberg & S.

B. Kaufman (Eds.), The Cambridge Handbook of Intelligence (1st ed., pp. 771–783).

Cambridge University Press. [Link]

Kavé, G., & Halamish, V. (2015). Doubly blessed: Older adults know more vocabulary and

know better what they know. Psychology and Aging, 30(1), 68–73.

[Link]

200
Kavé Gitit & Yafé Ronit. (2014). Performance of Younger and Older Adults on Tests of Word

Knowledge and Word Retrieval: Independence or Interdependence of Skills?

American Journal of Speech-Language Pathology, 23(1), 36–45.

[Link]

Kellner, R., & Benedek, M. (2017). The role of creative potential and intelligence for humor

production. Psychology of Aesthetics, Creativity, and the Arts, 11(1), 52–58.

[Link]

Kemper, S., Schmalzried, R., Hoffman, L., & Herman, R. (2010). Aging and the vulnerability

of speech to dual task demands. Psychology and Aging, 25(4), 949–962.

[Link]

Kenett, Y. N. (2018). Going the Extra Creative Mile: The Role of Semantic Distance in

Creativity – Theory, Research, and Measurement. In R. E. Jung & O. Vartanian (Eds.),

The Cambridge Handbook of the Neuroscience of Creativity (1st ed., pp. 233–248).

Cambridge University Press. [Link]

Kenett, Y. N., Anaki, D., & Faust, M. (2014). Investigating the structure of semantic networks

in low and high creative persons. Frontiers in Human Neuroscience, 8.

[Link]

Kenett, Y. N., & Austerweil, J. L. (2016). Examining Search Processes in Low and High

Creative Individuals with Random Walks.

Kenett, Y. N., & Faust, M. (2019). A Semantic Network Cartography of the Creative Mind.

Trends in Cognitive Sciences, 23(4), 271–274.

[Link]

Kenett, Y. N., Levy, O., Kenett, D. Y., Stanley, H. E., Faust, M., & Havlin, S. (2018).

Flexibility of thought in high creative individuals represented by percolation analysis.

201
 Proceedings of the National Academy of Sciences, 115(5), 867–872.

[Link]

Kenett, Y. N., Medaglia, J. D., Beaty, R. E., Chen, Q., Betzel, R. F., Thompson-Schill, S. L.,

& Qiu, J. (2018). Driving the brain towards creativity and intelligence: A network

control theory analysis. Neuropsychologia, 118, 79–90.

[Link]

Kiefer, M., & Pulvermüller, F. (2012). Conceptual representations in mind and brain:

Theoretical developments, current evidence and future directions. Cortex, 48(7), 805–

825. [Link]

Kim, H. (2016). Default network activation during episodic and semantic memory retrieval:

A selective meta-analytic comparison. Neuropsychologia, 80, 35–46.

[Link]

Kim, K. H. (2005). Can Only Intelligent People Be Creative? A Meta-Analysis.

Kintz, S., Fergadiotis, G., & Wright, H. H. (2016). Aging effects on discourse production.

Cognition, Language and Aging, 81–106.

Kleinmintz, O. M., Abecasis, D., Tauber, A., Geva, A., Chistyakov, A. V., Kreinin, I., Klein,

E., & Shamay-Tsoory, S. G. (2018a). Participation of the left inferior frontal gyrus in

human originality. Brain Structure and Function, 223(1), 329–341.

[Link]

Kleinmintz, O. M., Abecasis, D., Tauber, A., Geva, A., Chistyakov, A. V., Kreinin, I., Klein,

E., & Shamay-Tsoory, S. G. (2018b). Participation of the left inferior frontal gyrus in

human originality. Brain Structure and Function, 223(1), 329–341.

[Link]

Kleinmintz, O. M., Ivancovsky, T., & Shamay-Tsoory, S. G. (2019). The two-fold model of

creativity: The neural underpinnings of the generation and evaluation of creative

202
 ideas. Current Opinion in Behavioral Sciences, 27, 131–138.

[Link]

Krieger-Redwood, K. (n.d.). Creativity in Verbal Associations is Linked to Semantic Control.

Krieger-Redwood, K., Steward, A., Gao, Z., Wang, X., Halai, A., Smallwood, J., & Jefferies,

E. (2023). Creativity in verbal associations is linked to semantic control. Cerebral

Cortex, 33(9), 5135–5147. [Link]

Kröger, S., Rutter, B., Stark, R., Windmann, S., Hermann, C., & Abraham, A. (2012). Using a

shoe as a plant pot: Neural correlates of passive conceptual expansion. Brain

Research, 1430, 52–61. [Link]

Krug, S. E. (1973). Psychometric properties of the culture fair scales: Reliability and validity.

Technical Supplement for the Culture Fair Intelligence Tests, Scales 2 And, 3.

Kuhnke, P., Chapman, C. A., Cheung, V. K. M., Turker, S., Graessner, A., Martin, S.,

Williams, K. A., & Hartwigsen, G. (2023). The role of the angular gyrus in semantic

cognition: A synthesis of five functional neuroimaging studies. Brain Structure and

Function, 228(1), 273–291. [Link]

Kundu, P., Voon, V., Balchandani, P., Lombardo, M. V., Poser, B. A., & Bandettini, P. A.

(2017). Multi-echo fMRI: a review of applications in fMRI denoising and analysis of

BOLD signals. Neuroimage, 154, 59–80.

Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017). lmerTest package: Tests in

linear mixed effects models. Journal of Statistical Software, 82(13).

Lambon Ralph, M. A., Ehsan, S., Baker, G. A., & Rogers, T. T. (2012). Semantic memory is

impaired in patients with unilateral anterior temporal lobe resection for temporal lobe

epilepsy. Brain, 135(1), 242–258. [Link]

203
Lambon Ralph, M. A., Jefferies, E., Patterson, K., & Rogers, T. T. (2017). The neural and

computational bases of semantic cognition. Nature Reviews Neuroscience, 18(1), 42–

55. [Link]

Lambon Ralph, M. A. L., Jefferies, E., Patterson, K., & Rogers, T. T. (2017). The neural and

computational bases of semantic cognition. Nature Reviews Neuroscience, 18(1), 42–

55. [Link]

Lambon Ralph, M. A., Lowe, C., & Rogers, T. T. (2006). Neural basis of category-specific

semantic deficits for living things: Evidence from semantic dementia, HSVE and a

neural network model. Brain, 130(4), 1127–1137.

[Link]

Lambon Ralph, M. A., & Pobric, G. (2009). Conceptual knowledge is underpinned by the

temporal pole bilaterally: Novel data from rTMS.

Lambon Ralph, M. A., Sage, K., Jones, R. W., & Mayberry, E. J. (2010a). Coherent concepts

are computed in the anterior temporal lobes. Proceedings of the National Academy of

Sciences, 107(6), 2717–2722. [Link]

Lambon Ralph, M. A., Sage, K., Jones, R. W., & Mayberry, E. J. (2010b). Coherent concepts

are computed in the anterior temporal lobes. Proceedings of the National Academy of

Sciences, 107(6), 2717–2722. [Link]

Landauer, T. K., & Dumais, S. T. (1997). A solution to Plato’s problem: The latent semantic

analysis theory of acquisition, induction, and representation of knowledge.

Psychological Review, 104(2), 211.

Lanzoni, L., Thompson, H., Beintari, D., Berwick, K., Demnitz-King, H., Raspin, H., Taha,

M., Stampacchia, S., Smallwood, J., & Jefferies, E. (2019). Emotion and location cues

bias conceptual retrieval in people with deficient semantic control. Neuropsychologia,

131, 294–305. [Link]

204
Lee, C. S., Huggins, A. C., & Therriault, D. J. (2014). A measure of creativity or intelligence?

Examining internal and external structure validity evidence of the Remote Associates

Test. Psychology of Aesthetics, Creativity, and the Arts, 8(4), 446–460.

[Link]

Lee, C. S., & Therriault, D. J. (2013). The cognitive underpinnings of creative thought: A

latent variable analysis exploring the roles of intelligence and working memory in

three creative thinking processes. Intelligence, 41(5), 306–320.

[Link]

Leon, S. A., Altmann, L. J. P., Abrams, L., Gonzalez Rothi, L. J., & Heilman, K. M. (2014).

Divergent Task Performance in Older Adults: Declarative Memory or Creative

Potential? Creativity Research Journal, 26(1), 21–29.

[Link]

Lin, Q., Liu, J., Zheng, Y., Liang, H., & Calhoun, V. D. (2010). Semiblind spatial ICA of

fMRI using spatial constraints. Human Brain Mapping, 31(7), 1076–1088.

[Link]

Liu, S., Erkkinen, M. G., Healey, M. L., Xu, Y., Swett, K. E., Chow, H. M., & Braun, A. R.

(2015). Brain activity and connectivity during poetry composition: Toward a

multidimensional model of the creative process. Human Brain Mapping, 36(9), 3351–

3372. [Link]

Luchini, S., & Beaty, R. E. (2023). Brain networks of creative cognition. In Handbook of

Organizational Creativity (pp. 195–207). Elsevier. [Link]

323-91840-4.00021-9

Luchini, S., Kenett, Y. N., Zeitlen, D. C., Christensen, A. P., Ellis, D. M., Brewer, G. A., &

Beaty, R. E. (2023). Convergent thinking and insight problem solving relate to

205
 semantic memory network structure. Thinking Skills and Creativity, 48, 101277.

[Link]

Lüdecke, D. (2018). ggeffects: Tidy Data Frames of Marginal Effects from Regression

Models. Journal of Open Source Software, 3(26), 772.

[Link]

Luria, A. R. (1973). Neuropsychological studies in the USSR. A review. I. Proceedings of the

National Academy of Sciences, 70(3), 959-964.

MacPherson, S. E., Phillips, L. H., & Della Sala, S. (2002). Age, executive function and

social decision making: A dorsolateral prefrontal theory of cognitive aging.

Psychology and Aging, 17(4), 598–609. [Link]

Madore, K. P., Addis, D. R., & Schacter, D. L. (2015). Creativity and Memory: Effects of an

Episodic-Specificity Induction on Divergent Thinking. Psychological Science, 26(9),

1461–1468. [Link]

Madore, K. P., Jing, H. G., & Schacter, D. L. (2016). Divergent creative thinking in young

and older adults: Extending the effects of an episodic specificity induction. Memory &

Cognition, 44(6), 974–988. [Link]

Madore, K. P., Thakral, P. P., Beaty, R. E., Addis, D. R., & Schacter, D. L. (2019). Neural

Mechanisms of Episodic Retrieval Support Divergent Creative Thinking. Cerebral

Cortex, 29(1), 150–166. [Link]

Maldonado, T., Orr, J. M., Goen, J. R. M., & Bernard, J. A. (2020). Age Differences in the

Subcomponents of Executive Functioning. The Journals of Gerontology: Series B,

75(6), e31–e55. [Link]

Margulies, D. S., Ghosh, S. S., Goulas, A., Falkiewicz, M., Huntenburg, J. M., Langs, G.,

Bezgin, G., Eickhoff, S. B., Castellanos, F. X., Petrides, M., Jefferies, E., &

Smallwood, J. (2016). Situating the default-mode network along a principal gradient

206
 of macroscale cortical organization. Proceedings of the National Academy of

Sciences, 113(44), 12574–12579. [Link]

Marini, A. (2012). Characteristics of narrative discourse processing after damage to the right

hemisphere. Seminars in Speech and Language, 33(01), 68–78.

Marini, A., & Andreetta, S. (2016). Age-related effects on language production. Cognition,

Language and Aging. Amsterdam: John Benjamins, 55–79.

Marini, A., Andreetta, S., Del Tin, S., & Carlomagno, S. (2011). A multi-level approach to the

analysis of narrative language in aphasia. Aphasiology, 25(11), 1372–1392.

[Link]

Marini, A., Boewe, A., Caltagirone, C., & Carlomagno, S. (2005). Age-related differences in

the production of textual descriptions. Journal of Psycholinguistic Research, 34, 439–

463.

Marinkovic, K., Dhond, R. P., Dale, A. M., Glessner, M., Carr, V., & Halgren, E. (2003).

Spatiotemporal Dynamics of Modality-Specific and Supramodal Word Processing.

Neuron, 38(3), 487–497. [Link]

Marko, M., Michalko, D., Dragašek, J., Vančová, Z., Jarčušková, D., & Riečanský, I. (2022).

Assessment of Automatic and Controlled Retrieval Using Verbal Fluency Tasks.

Marko, M., Michalko, D., & Riečanský, I. (2019). Remote associates test: An empirical proof

of concept. Behavior Research Methods, 51(6), 2700–2711.

[Link]

Marko, M., & Riečanský, I. (2023). Differential effects of executive load on automatic versus

controlled semantic memory retrieval. Memory & Cognition, 51(5), 1145–1158.

[Link]

Marron, T. R., Lerner, Y., Berant, E., Kinreich, S., Shapira-Lichter, I., Hendler, T., & Faust,

M. (2018). Chain free association, creativity, and the default mode network.

207
 Neuropsychologia, 118, 40–58.

[Link]

Martin, A. (2007). The representation of object concepts in the brain. Annu. Rev. Psychol.,

58(1), 25–45.

Massimiliano, P. (2015). The effects of age on divergent thinking and creative objects

production: A cross-sectional study. High Ability Studies, 26(1), 93–104.

[Link]

Masson, H. L., Op De Beeck, H., & Boets, B. (2020). Reduced task-dependent modulation of

functional network architecture for positive versus negative affective touch processing

in autism spectrum disorders. NeuroImage, 219, 117009.

[Link]

Mastria, S., Agnoli, S., Zanon, M., Acar, S., Runco, M. A., & Corazza, G. E. (2021).

Clustering and switching in divergent thinking: Neurophysiological correlates

underlying flexibility during idea generation. Neuropsychologia, 158, 107890.

[Link]

Mayseless, N., Aharon-Peretz, J., & Shamay-Tsoory, S. (2014). Unleashing creativity: The

role of left temporoparietal regions in evaluating and inhibiting the generation of

creative ideas. Neuropsychologia, 64, 157–168.

[Link]

Mayseless, N., Eran, A., & Shamay-Tsoory, S. G. (2015). Generating original ideas: The

neural underpinning of originality. NeuroImage, 116, 232–239.

[Link]

McMillan, R. L., Kaufman, S. B., & Singer, J. L. (2013). Ode to positive constructive

daydreaming. Frontiers in Psychology, 4. [Link]

208
Mednick, S. (1962). The associative basis of the creative process. Psychological Review, 69,

220–232. [Link]

Mekern, V. N., Sjoerds, Z., & Hommel, B. (2019). How metacontrol biases and adaptivity

impact performance in cognitive search tasks. Cognition, 182, 251–259.

[Link]

Meteyard, L., Cuadrado, S. R., Bahrami, B., & Vigliocco, G. (2012). Coming of age: A

review of embodiment and the neuroscience of semantics. Cortex, 48(7), 788–804.

[Link]

Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient Estimation of Word

Representations in Vector Space (arXiv:1301.3781). arXiv.

[Link]

Mion, M., Patterson, K., Acosta-Cabronero, J., Pengas, G., Izquierdo-Garcia, D., Hong, Y. T.,

Fryer, T. D., Williams, G. B., Hodges, J. R., & Nestor, P. J. (2010). What the left and

right anterior fusiform gyri tell us about semantic memory. Brain, 133(11), 3256–

3268. [Link]

Miroshnik, K. G., Forthmann, B., Karwowski, M., & Benedek, M. (2023). The relationship of

divergent thinking with broad retrieval ability and processing speed: A meta-analysis.

Intelligence, 98, 101739. [Link]

Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D.

(2000). The Unity and Diversity of Executive Functions and Their Contributions to

Complex “Frontal Lobe” Tasks: A Latent Variable Analysis. Cognitive Psychology,

41(1), 49–100. [Link]

Moeller, S., Yacoub, E., Olman, C. A., Auerbach, E., Strupp, J., Harel, N., & Uğurbil, K.

(2010). Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial

209
 parallel imaging with application to high spatial and temporal whole-brain fMRI.

Magnetic Resonance in Medicine, 63(5), 1144–1153.

Mollo, G., Cornelissen, P. L., Millman, R. E., Ellis, A. W., & Jefferies, E. (2017). Oscillatory

Dynamics Supporting Semantic Cognition: MEG Evidence for the Contribution of the

Anterior Temporal Lobe Hub and Modality-Specific Spokes. PLOS ONE, 12(1),

e0169269. [Link]

Montefinese, M., Hallam, G., Stampacchia, S., Thompson, H. E., & Jefferies, E. (2021).

Deficits of semantic control disproportionately affect low-relevance conceptual

features: Evidence from semantic aphasia. Aphasiology, 35(11), 1448–1462.

[Link]

Morales, M., Patel, T., Tamm, A., Pickering, M. J., & Hoffman, P. (2022). Similar neural

networks respond to coherence during comprehension and production of discourse.

Cerebral Cortex, 32(19), 4317–4330. [Link]

Niendam, T. A., Laird, A. R., Ray, K. L., Dean, Y. M., Glahn, D. C., & Carter, C. S. (2012).

Meta-analytic evidence for a superordinate cognitive control network subserving

diverse executive functions. Cognitive, Affective, & Behavioral Neuroscience, 12(2),

241–268. [Link]

Nijstad, B. A., De Dreu, C. K. W., Rietzschel, E. F., & Baas, M. (2010). The dual pathway to

creativity model: Creative ideation as a function of flexibility and persistence.

European Review of Social Psychology, 21(1), 34–77.

[Link]

Noonan, K. A., Jefferies, E., Corbett, F., & Lambon Ralph, M. A. (2010). Elucidating the

Nature of Deregulated Semantic Cognition in Semantic Aphasia: Evidence for the

Roles of Prefrontal and Temporo-parietal Cortices. Journal of Cognitive

Neuroscience, 22(7), 1597–1613. [Link]

210
Noonan, K. A., Jefferies, E., Visser, M., & Lambon Ralph, M. A. (2013). Going beyond

Inferior Prefrontal Involvement in Semantic Control: Evidence for the Additional

Contribution of Dorsal Angular Gyrus and Posterior Middle Temporal Cortex. Journal

of Cognitive Neuroscience, 25(11), 1824–1850. [Link]

Nusbaum, E. C., & Silvia, P. J. (2011). Are intelligence and creativity really so

different?☆Fluid intelligence, executive processes, and strategy use in divergent

thinking. Intelligence, 39(1), 36–45. [Link]

Nusbaum, E. C., Silvia, P. J., & Beaty, R. E. (2014). Ready, set, create: What instructing

people to “be creative” reveals about the meaning and mechanisms of divergent

thinking. Psychology of Aesthetics, Creativity, and the Arts, 8(4), 423–432.

[Link]

Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh inventory.

Neuropsychologia, 9(1), 97–113.

Oleynick, V. C., DeYoung, C. G., Hyde, E., Kaufman, S. B., Beaty, R. E., & Silvia, P. J.

(2017). Openness/Intellect: The Core of the Creative Personality. In G. J. Feist, R.

Reiter-Palmon, & J. C. Kaufman (Eds.), The Cambridge Handbook of Creativity and

Personality Research (1st ed., pp. 9–27). Cambridge University Press.

[Link]

Organisciak, P., Selcuk Acar, Dumas, D., & Berthiaume, K. (2022). Beyond Semantic

Distance: Automated Scoring of Divergent Thinking Greatly Improves with Large

Language Models. [Link]

Orwig, W., Beaty, R. E., Benedek, M., & Schacter, D. L. (2024). Creative Evaluation: The

Role of Memory in Novelty & Effectiveness Judgements. Creativity Research

Journal, 1–9. [Link]

211
Orwig, W., Diez, I., Vannini, P., Beaty, R., & Sepulcre, J. (2021). Creative Connections:

Computational Semantic Distance Captures Individual Creativity and Resting-State

Functional Connectivity. Journal of Cognitive Neuroscience, 33(3), 499–509.

[Link]

Orwig, W., Edenbaum, E. R., Greene, J. D., & Schacter, D. L. (2024). The Language of

Creativity: Evidence from Humans and Large Language Models. The Journal of

Creative Behavior, 58(1), 128–136. [Link]

Ovando-Tellez, M., Benedek, M., Kenett, Y. N., Hills, T., Bouanane, S., Bernard, M., Belo, J.,

Bieth, T., & Volle, E. (2022). An investigation of the cognitive and neural correlates of

semantic memory search related to creative ability. Communications Biology, 5(1),

604. [Link]

Palmiero, M., Fusi, G., Crepaldi, M., Borsa, V. M., & Rusconi, M. L. (2022). Divergent

thinking and the core executive functions: A state-of-the-art review. Cognitive

Processing, 23(3), 341–366. [Link]

Palmiero, M., Nori, R., & Piccardi, L. (2017). Verbal and visual divergent thinking in aging.

Experimental Brain Research, 235(4), 1021–1029. [Link]

016-4857-4

Park, D. C., Lautenschlager, G., Hedden, T., Davidson, N. S., Smith, A. D., & Smith, P. K.

(2002). Models of visuospatial and verbal memory across the adult life span.

Psychology and Aging, 17, 299–320. [Link]

Patel, T., MacPherson, S. E., & Hoffman, P. (n.d.). Investigating age related differences in

semantic control mechanisms involved in creative cognition.

[Link]

212
Patel, T., Morales, M., Pickering, M. J., & Hoffman, P. (2023). A common neural code for

meaning in discourse production and comprehension. NeuroImage, 279, 120295.

[Link]

Patterson, K., & Lambon Ralph, M. A. (2016). The Hub-and-Spoke Hypothesis of Semantic

Memory. In Neurobiology of Language (pp. 765–775). Elsevier.

[Link]

Patterson, K., Nestor, P. J., & Rogers, T. T. (2007). Where do you know what you know? The

representation of semantic knowledge in the human brain. Nature Reviews

Neuroscience, 8(12), 976–987. [Link]

Patterson, K., Ralph, M. A. L., Jefferies, E., Woollams, A., Jones, R., Hodges, J. R., &

Rogers, T. T. (2006). “Presemantic” Cognition in Semantic Dementia: Six Deficits in

Search of an Explanation. Journal of Cognitive Neuroscience, 18(2), 169–183.

[Link]

Paulin, T., Roquet, D., Kenett, Y. N., Savage, G., & Irish, M. (2020). The effect of semantic

memory degeneration on creative thinking: A voxel-based morphometry analysis.

NeuroImage, 220, 117073. [Link]

Pedersen, T. L. (2023). patchwork: The Composer of Plots. [Link]

[Link]/package=patchwork

Peelen, M. V., & Caramazza, A. (2012). Conceptual Object Representations in Human

Anterior Temporal Cortex. The Journal of Neuroscience, 32(45), 15728–15736.

[Link]

Peterson, D. R., & Pattie, M. W. (2024). Think Outside and Inside the Box:The Role of Dual-

Pathway Divergent Thinking in Creative Idea Generation. Creativity Research

Journal, 36(2), 272–290. [Link]

213
Pinho, A. L., Ullén, F., Castelo-Branco, M., Fransson, P., & De Manzano, Ö. (2015).

Addressing a Paradox: Dual Strategies for Creative Performance in Introspective and

Extrospective Networks. Cerebral Cortex, 26(7), 3052–3063.

[Link]

Plaut, D. C. (2002). Graded modality-specific specialisation in semantics: A computational

account of optic aphasia. Cognitive Neuropsychology, 19(7), 603–639.

Pobric, G., Jefferies, E., & Lambon Ralph, M. A. (2010a). Amodal semantic representations

depend on both anterior temporal lobes: Evidence from repetitive transcranial

magnetic stimulation. Neuropsychologia, 48(5), 1336–1342.

[Link]

Pobric, G., Jefferies, E., & Lambon Ralph, M. A. (2010b). Category-Specific versus

Category-General Semantic Impairment Induced by Transcranial Magnetic

Stimulation. Current Biology, 20(10), 964–968.

[Link]

Pobric, G., Jefferies, E., & Ralph, M. A. L. (2007). Anterior temporal lobes mediate semantic

representation: Mimicking semantic dementia by using rTMS in normal participants.

Proceedings of the National Academy of Sciences, 104(50), 20137–20141.

[Link]

Pobric, G., Lambon Ralph, M. A., & Jefferies, E. (2009). The role of the anterior temporal

lobes in the comprehension of concrete and abstract words: rTMS evidence. Cortex,

45(9), 1104–1110. [Link]

Pulvermüller, F. (2005). Brain mechanisms linking language and action. Nature Reviews

Neuroscience, 6(7), 576–582.

R Core Team. (2023). R: A Language and Environment for Statistical Computing. R

Foundation for Statistical Computing. [Link]

214
Raaijmakers, J. G., & Shiffrin, R. M. (1981). Search of associative memory. Psychological

Review, 88(2), 93.

Radel, R., Davranche, K., Fournier, M., & Dietrich, A. (2015). The role of (dis)inhibition in

creativity: Decreased inhibition improves idea generation. Cognition, 134, 110–120.

[Link]

Raichle, M. E., & Snyder, A. Z. (2007). A default mode of brain function: A brief history of

an evolving idea. NeuroImage, 37(4), 1083–1090.

[Link]

Ram, K., & Wickham, H. (2023). wesanderson: A Wes Anderson Palette Generator.

[Link]

Rastelli, C., Greco, A., Finocchiaro, C., Penazzi, G., Braun, C., & De Pisapia, N. (2024).

Neural dynamics of semantic control underlying generative storytelling.

[Link]

Raven, J. C., Court, J. H., & Raven, J. E. (1989). Standard progressive matrices. Australian

Council for Educational Research Limited Camberwell, Australia.

Raven, J., & Raven, J. (2003). Raven Progressive Matrices. In Handbook of nonverbal

assessment. (pp. 223–237). Kluwer Academic/Plenum Publishers.

[Link]

Reilly, J., Shain, C., Borghesani, V., Kuhnke, P., Vigliocco, G., Peelle, J. E., Mahon, B.,

Buxbaum, L., Majid, A., Brysbaert, M., Borghi, A. M., De Deyne, S., Dove, G.,

Papeo, L., Pexman, P. M., Poeppel, D., Lupyan, G., Boggio, P., Hickock, G., …

Vinson, D. (2023). What we mean when we say semantic: A Consensus statement on

the nomenclature of semantic memory. [Link]

215
Reiter-Palmon, R., Forthmann, B., & Barbot, B. (2019). Scoring divergent thinking tests: A

review and systematic framework. Psychology of Aesthetics, Creativity, and the Arts,

13(2), 144–152. [Link]

Reiter-Palmon, R., Wigert, B., & Vreede, T. de. (2012). Chapter 13—Team Creativity and

Innovation: The Effect of Group Composition, Social Processes, and Cognition. In M.

D. Mumford (Ed.), Handbook of Organizational Creativity (pp. 295–326). Academic

Press. [Link]

Rey-Mermet, A., & Gade, M. (2018). Inhibition in aging: What is preserved? What declines?

A meta-analysis. Psychonomic Bulletin & Review, 25(5), 1695–1716.

[Link]

Rhodes, M. (1961). An Analysis of Creativity. The Phi Delta Kappan, 42(7), 305–310.

JSTOR.

Rice, G. E., Lambon Ralph, M. A., & Hoffman, P. (2015). The Roles of Left Versus Right

Anterior Temporal Lobes in Conceptual Knowledge: An ALE Meta-analysis of 97

Functional Neuroimaging Studies. Cerebral Cortex, 25(11), 4374–4391.

[Link]

Richie, R., Aka, A., & Bhatia, S. (2023). Free association in a neural network. Psychological

Review, 130(5), 1360–1382. [Link]

Rogers, R. D., & Monsell, S. (1995). Costs of a predictible switch between simple cognitive

tasks. Journal of Experimental Psychology: General, 124(2), 207.

Rogers, T. T., Hocking, J., Noppeney, U., Mechelli, A., Gorno-Tempini, M. L., Patterson, K.,

& Price, C. J. (2006). Anterior temporal cortex and semantic memory: Reconciling

findings from neuropsychology and functional imaging. Cognitive, Affective, &

Behavioral Neuroscience, 6(3), 201–213.

216
Rogers, T. T., Lambon Ralph, M. A., Garrard, P., Bozeat, S., McClelland, J. L., Hodges, J. R.,

& Patterson, K. (2004). Structure and Deterioration of Semantic Memory: A

Neuropsychological and Computational Investigation. Psychological Review, 111(1),

205–235. [Link]

Rogers, T. T., Patterson, K., Jefferies, E., & Lambon Ralph, M. A. (2015a). Disorders of

representation and control in semantic cognition: Effects of familiarity, typicality, and

specificity. Neuropsychologia, 76, 220–239.

[Link]

Rogers, T. T., Patterson, K., Jefferies, E., & Lambon Ralph, M. A. (2015b). Disorders of

representation and control in semantic cognition: Effects of familiarity, typicality, and

specificity. Neuropsychologia, 76, 220–239.

[Link]

Rossmann, E., & Fink, A. (2010). Do creative people use shorter associative pathways?

Personality and Individual Differences, 49(8), 891–895.

[Link]

Runco, M. A. (2008). Commentary: Divergent thinking is not synonymous with creativity.

Psychology of Aesthetics, Creativity, and the Arts, 2(2), 93–96.

[Link]

Runco, M. A., & Acar, S. (2012). Divergent Thinking as an Indicator of Creative Potential.

Creativity Research Journal, 24(1), 66–75.

[Link]

Runco, M. A., Dow, G., & Smith, W. R. (2006). Information, Experience, and Divergent

Thinking: An Empirical Test. Creativity Research Journal, 18(3), 269–277.

[Link]

217
Runco, M. A., & Jaeger, G. J. (2012). The Standard Definition of Creativity. Creativity

Research Journal, 24(1), 92–96. [Link]

Ryan, J. J., Sattler, J. M., & Lopez, S. J. (2000). Age effects on Wechsler Adult Intelligence

Scale-III subtests. Archives of Clinical Neuropsychology, 15, 311–317.

[Link]

Salthouse, T. A. (1992). Mechanisms of age–cognition relations in adulthood. (pp. vi, 132).

Lawrence Erlbaum Associates, Inc.

Salthouse, T. A. (2000). Aging and measures of processing speed. Biological Psychology,

54(1–3), 35–54. [Link]

Salthouse, T. A. (2004). Localizing age-related individual differences in a hierarchical

structure. Intelligence, 32(6), 541–561. [Link]

Salthouse, T. A. (2009). When does age-related cognitive decline begin? Neurobiology of

Aging, 30(4), 507–514. [Link]

Schacter, D. L., Addis, D. R., Hassabis, D., Martin, V. C., Spreng, R. N., & Szpunar, K. K.

(2012). The Future of Memory: Remembering, Imagining, and the Brain. Neuron,

76(4), 677–694. [Link]

Schneider, W. J., & McGrew, K. S. (2018). The Cattell-Horn-Carroll theory of cognitive

abilities. Contemporary Intellectual Assessment: Theories, Tests, and Issues, 73–163.

Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., Reiss, A. L.,

& Greicius, M. D. (2007). Dissociable Intrinsic Connectivity Networks for Salience

Processing and Executive Control. The Journal of Neuroscience, 27(9), 2349–2356.

[Link]

Sen, S., & Yörük, S. (2023). A Reliability Generalization Meta‐Analysis of the Kaufman

Domains of Creativity Scale. The Journal of Creative Behavior, 57(4), 812–837.

[Link]

218
Shamay-Tsoory, S. G., Adler, N., Aharon-Peretz, J., Perry, D., & Mayseless, N. (2011). The

origins of originality: The neural bases of creative thinking and originality.

Neuropsychologia, 49(2), 178–185.

[Link]

Shi, L., Sun, J., Xia, Y., Ren, Z., Chen, Q., Wei, D., Yang, W., & Qiu, J. (2018). Large-scale

brain network connectivity underlying creativity in resting-state and task fMRI:

Cooperation between default network and frontal-parietal network. Biological

Psychology, 135, 102–111. [Link]

Shofty, B., Gonen, T., Bergmann, E., Mayseless, N., Korn, A., Shamay-Tsoory, S., Grossman,

R., Jalon, I., Kahn, I., & Ram, Z. (2022). The default network is causally linked to

creative thinking. Molecular Psychiatry, 27(3), 1848–1854.

[Link]

Silvia, P. J. (2015). Intelligence and Creativity Are Pretty Similar After All. Educational

Psychology Review, 27(4), 599–606. [Link]

Silvia, P. J., & Beaty, R. E. (2012). Making creative metaphors: The importance of fluid

intelligence for creative thought. Intelligence, 40(4), 343–351.

[Link]

Silvia, P. J., Beaty, R. E., & Nusbaum, E. C. (2013). Verbal fluency and creativity: General

and specific contributions of broad retrieval ability (Gr) factors to divergent thinking.

Intelligence, 41(5), 328–340. [Link]

Silvia, P. J., Wigert, B., Reiter-Palmon, R., & Kaufman, J. C. (2012). Assessing creativity

with self-report scales: A review and empirical evaluation. Psychology of Aesthetics,

Creativity, and the Arts, 6(1), 19–34. [Link]

Silvia, P. J., Winterstein, B. P., Willse, J. T., Barona, C. M., Cram, J. T., Hess, K. I., Martinez,

J. L., & Richard, C. A. (2008). Assessing creativity with divergent thinking tasks:

219
 Exploring the reliability and validity of new subjective scoring methods. Psychology

of Aesthetics, Creativity, and the Arts, 2(2), 68–85. [Link]

3896.2.2.68

Simonton, D. K. (1999). Talent and its development: An emergenic and epigenetic model.

Psychological Review, 106, 435–457. [Link]

Simonton, D. K. (2012). Taking the U.S. Patent Office Criteria Seriously: A Quantitative

Three-Criterion Creativity Definition and Its Implications. Creativity Research

Journal, 24(2–3), 97–106. [Link]

Sio, U. N., Kotovsky, K., & Cagan, J. (2022). Determinants of creative thinking: The effect of

task characteristics in solving remote associate test problems. Thinking & Reasoning,

28(2), 163–192. [Link]

Sliwinski, M., & Buschke, H. (1999). Cross-Sectional and Longitudinal Relationships Among

Age, Cognition, and Processing Speed.

Smallwood, J., Bernhardt, B. C., Leech, R., Bzdok, D., Jefferies, E., & Margulies, D. S.

(2021). The default mode network in cognition: A topographical perspective. Nature

Reviews Neuroscience, 22(8), 503–513. [Link]

Smallwood, J., Brown, K., Baird, B., & Schooler, J. W. (2012). Cooperation between the

default mode network and the frontal–parietal network in the production of an internal

train of thought. Brain Research, 1428, 60–70.

[Link]

Smallwood, J., & Schooler, J. W. (2015). The Science of Mind Wandering: Empirically

Navigating the Stream of Consciousness. Annual Review of Psychology, 66(1), 487–

518. [Link]

220
Smith, K. A., Huber, D. E., & Vul, E. (2013). Multiply-constrained semantic search in the

Remote Associates Test. Cognition, 128(1), 64–75.

[Link]

Smith, K. A., & Vul, E. (2015). The Role of Sequential Dependence in Creative Semantic

Search. Topics in Cognitive Science, 7(3), 543–546.

[Link]

Snowden, J., Bathgate, D., Varma, A., Blackshaw, A., Gibbons, Z., & Neary, D. (2001).

Distinct behavioural profiles in frontotemporal dementia and semantic dementia.

Journal of Neurology, Neurosurgery & Psychiatry, 70(3), 323–332.

Soveri, A., Lehtonen, M., Karlsson, L. C., Lukasik, K., Antfolk, J., & Laine, M. (2018). Test–

retest reliability of five frequently used executive tasks in healthy adults. Applied

Neuropsychology: Adult, 25(2), 155–165.

[Link]

Sowden, P. T., Pringle, A., & Gabora, L. (2015). The shifting sands of creative thinking:

Connections to dual-process theory. Thinking & Reasoning, 21(1), 40–60.

[Link]

Spreng, R. N., Mar, R. A., & Kim, A. S. N. (2009). The Common Neural Basis of

Autobiographical Memory, Prospection, Navigation, Theory of Mind, and the Default

Mode: A Quantitative Meta-analysis. Journal of Cognitive Neuroscience, 21(3), 489–

510. [Link]

Spreng, R. N., Stevens, W. D., Chamberlain, J. P., Gilmore, A. W., & Schacter, D. L. (2010).

Default network activity, coupled with the frontoparietal control network, supports

goal-directed cognition. NeuroImage, 53(1), 303–317.

[Link]

221
Spreng, R. N., & Turner, G. R. (2019). The Shifting Architecture of Cognition and Brain

Function in Older Adulthood.

Stein, M. I. (1953). Creativity and Culture. The Journal of Psychology, 36(2), 311–322.

[Link]

Stephens, G. J., Silbert, L. J., & Hasson, U. (2010). Speaker–listener neural coupling

underlies successful communication. Proceedings of the National Academy of

Sciences, 107(32), 14425–14430. [Link]

Takeuchi, H., Taki, Y., Hashizume, H., Sassa, Y., Nagase, T., Nouchi, R., & Kawashima, R.

(2012). The Association between Resting Functional Connectivity and Creativity.

Cerebral Cortex, 22(12), 2921–2929. [Link]

Teige, C., Cornelissen, P. L., Mollo, G., Gonzalez Alam, T. R. D. J., McCarty, K., Smallwood,

J., & Jefferies, E. (2019). Dissociations in semantic cognition: Oscillatory evidence

for opposing effects of semantic control and type of semantic relation in anterior and

posterior temporal cortex. Cortex, 120, 308–325.

[Link]

Thompson, H., Davey, J., Hoffman, P., Hallam, G., Kosinski, R., Howkins, S., Wooffindin,

E., Gabbitas, R., & Jefferies, E. (2017). Semantic control deficits impair

understanding of thematic relationships more than object identity. Neuropsychologia,

104, 113–125. [Link]

Thompson, H. E., Almaghyuli, A., Noonan, K. A., Barak, O., Lambon Ralph, M. A., &

Jefferies, E. (2018). The contribution of executive control to semantic cognition:

Convergent evidence from semantic aphasia and executive dysfunction. Journal of

Neuropsychology, 12(2), 312–340. [Link]

Thompson, H. E., Noonan, K. A., Halai, A. D., Hoffman, P., Stampacchia, S., Hallam, G.,

Rice, G. E., De Dios Perez, B., Lambon Ralph, M. A., & Jefferies, E. (2022). Damage

222
 to temporoparietal cortex is sufficient for impaired semantic control. Cortex, 156, 71–

85. [Link]

Thompson, H. E., Robson, H., Lambon Ralph, M. A., & Jefferies, E. (2015). Varieties of

semantic ‘access’ deficit in Wernicke’s aphasia and semantic aphasia. Brain, 138(12),

3776–3792. [Link]

Thompson-Schill, S. L. (2003). Neuroimaging studies of semantic memory: Inferring “how”

from “where”. Neuropsychologia, 41(3), 280–292. [Link]

3932(02)00161-6

Thompson-Schill, S. L., D’Esposito, M., Aguirre, G. K., & Farah, M. J. (1997). Role of left

inferior prefrontal cortex in retrieval of semantic knowledge: A reevaluation.

Proceedings of the National Academy of Sciences, 94(26), 14792–14797.

[Link]

Thompson-Schill, S. L., Jonides, J., Marshuetz, C., Smith, E. E., D’Esposito, M., Kan, I. P.,

& Knight, R. T. (2002). Effects of frontal lobe damage on interference effects in

working memory.

Todd, P. M., & Hills, T. T. (2020). Foraging in Mind. Current Directions in Psychological

Science, 29(3), 309–315. [Link]

Troyer, A. K. (2000). Normative Data for Clustering and Switching on Verbal Fluency Tasks.

Journal of Clinical and Experimental Neuropsychology, 22(3), 370–378.

[Link]

Troyer, A. K., Moscovitch, M., & Winocur, G. (1997). Clustering and switching as two

components of verbal fluency: Evidence from younger and older healthy adults.

Neuropsychology, 11(1), 138–146. [Link]

223
Trunk, D. L., & Abrams, L. (2009). Do younger and older adults’ communicative goals

influence off-topic speech in autobiographical narratives? Psychology and Aging,

24(2), 324–337. [Link]

Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review of Psychology,

53(1), 1–25.

Unsworth, N., Spillers, G. J., & Brewer, G. A. (2011). Variation in verbal fluency: A latent

variable analysis of clustering, switching, and overall performance. Quarterly Journal

of Experimental Psychology, 64(3), 447–466.

[Link]

Valba, O., Gorsky, A., Nechaev, S., & Tamm, M. (2021). Analysis of English free association

network reveals mechanisms of efficient solution of Remote Association Tests. PLOS

ONE, 16(4), e0248986. [Link]

Vartanian, O. (2009). Variable attention facilitates creative problem solving. Psychology of

Aesthetics, Creativity, and the Arts, 3(1), 57–59. [Link]

Vartanian, O., Beatty, E. L., Smith, I., Blackler, K., Lam, Q., & Forbes, S. (2018). One-way

traffic: The inferior frontal gyrus controls brain activation in the middle temporal

gyrus and inferior parietal lobule during divergent thinking. Neuropsychologia, 118,

68–78. [Link]

Vatansever, D., Smallwood, J., & Jefferies, E. (2021). Varying demands for cognitive control

reveals shared neural processes supporting semantic and episodic memory retrieval.

Nature Communications, 12(1), 2134. [Link]

Vaughan, L., Basak, C., Hartman, M., & Verhaeghen, P. (2008). Aging and Working Memory

Inside and Outside the Focus of Attention: Dissociations of Availability and

Accessibility. Aging, Neuropsychology, and Cognition, 15(6), 703–724.

[Link]

224
Velicer, W. F. (1976). Determining the number of components from the matrix of partial

correlations. Psychometrika, 41(3), 321–327. [Link]

Verhaeghen, P. (2003). Aging and vocabulary score: A meta-analysis. Psychology and Aging,

18, 332–339. [Link]

Verhaeghen, P., & Salthouse, T. A. (1997). Meta-Analyses of Age-Cognition Relations in

Adulthood: Estimates of Linear and Nonlinear Age Effects and Structural Models.

Visser, M., Jefferies, E., Embleton, K. V., & Lambon Ralph, M. A. (2012). Both the Middle

Temporal Gyrus and the Ventral Anterior Temporal Area Are Crucial for Multimodal

Semantic Processing: Distortion-corrected fMRI Evidence for a Double Gradient of

Information Convergence in the Temporal Lobes. Journal of Cognitive Neuroscience,

24(8), 1766–1778. [Link]

Visser, M., Jefferies, E., & Lambon Ralph, M. A. (2010). Semantic Processing in the Anterior

Temporal Lobes: A Meta-analysis of the Functional Neuroimaging Literature. Journal

of Cognitive Neuroscience, 22(6), 1083–1094.

[Link]

Visser, M., & Lambon Ralph, M. A. (2011). Differential Contributions of Bilateral Ventral

Anterior Temporal Lobe and Left Anterior Superior Temporal Gyrus to Semantic

Processes. Journal of Cognitive Neuroscience, 23(10), 3121–3131.

[Link]

Volle, E. (2018). Associative and Controlled Cognition in Divergent Thinking: Theoretical,

Experimental, Neuroimaging Evidence, and New Directions. In R. E. Jung & O.

Vartanian (Eds.), The Cambridge Handbook of the Neuroscience of Creativity (1st ed.,

pp. 333–360). Cambridge University Press.

[Link]

225
Wallach, M. A., & Kogan, N. (1965). A new look at the creativity-intelligence distinction1.

Journal of Personality, 33(3), 348–369. [Link]

6494.1965.tb01391.x

Wang, X., Bernhardt, B. C., Karapanagiotidis, T., De Caso, I., Gonzalez Alam, T. R. D. J.,

Cotter, Z., Smallwood, J., & Jefferies, E. (2018). The structural basis of semantic

control: Evidence from individual differences in cortical thickness. NeuroImage, 181,

480–489. [Link]

Wang, X., Margulies, D. S., Smallwood, J., & Jefferies, E. (2020). A gradient from long-term

memory to novel cognition: Transitions through default mode and executive cortex.

NeuroImage, 220, 117074. [Link]

Warrington, E. K. (1975). The Selective Impairment of Semantic Memory. Quarterly Journal

of Experimental Psychology, 27(4), 635–657.

[Link]

Warrington, E. K., & Shallice, T. (1979). Semantic access dyslexia. Brain: A Journal of

Neurology, 102(1), 43–63.

Wasylyshyn, C., Verhaeghen, P., & Sliwinski, M. J. (2011). Aging and task switching: A

meta-analysis. Psychology and Aging, 26(1), 15–20. [Link]

Wechsler, D. (1997). WAIS-III UK administration and scoring manual.

Wei, D., Yang, J., Li, W., Wang, K., Zhang, Q., & Qiu, J. (2014). Increased resting functional

connectivity of the medial prefrontal cortex in creativity by means of cognitive

stimulation. Cortex, 51, 92–102. [Link]

Weinstein, T. J., Ceh, S. M., Meinel, C., & Benedek, M. (2022). What’s Creative About

Sentences? A Computational Approach to Assessing Creativity in a Sentence

Generation Task. Creativity Research Journal, 34(4), 419–430.

[Link]

226
Weiss, S., Goecke, B., & Wilhelm, O. (2024). How Much Retrieval Ability Is in Originality?

The Journal of Creative Behavior, jocb.659. [Link]

Whiteside, D. M., Kealey, T., Semla, M., Luu, H., Rice, L., Basso, M. R., & Roper, B. (2016).

Verbal Fluency: Language or Executive Function Measure? Applied

Neuropsychology: Adult, 23(1), 29–34.

[Link]

Whitney, C., Kirk, M., O’Sullivan, J., Lambon Ralph, M. A., & Jefferies, E. (2011a). The

Neural Organization of Semantic Control: TMS Evidence for a Distributed Network

in Left Inferior Frontal and Posterior Middle Temporal Gyrus. Cerebral Cortex, 21(5),

1066–1075. [Link]

Whitney, C., Kirk, M., O’Sullivan, J., Lambon Ralph, M. A., & Jefferies, E. (2011b). The

Neural Organization of Semantic Control: TMS Evidence for a Distributed Network

in Left Inferior Frontal and Posterior Middle Temporal Gyrus. Cerebral Cortex, 21(5),

1066–1075. [Link]

Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L. D., François, R., Grolemund,

G., Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T. L., Miller, E., Bache, S.

M., Müller, K., Ooms, J., Robinson, D., Seidel, D. P., Spinu, V., … Yutani, H. (2019).

Welcome to the tidyverse. Journal of Open Source Software, 4(43), 1686.

[Link]

Wise, R. J. S., & Geranmayeh, F. (2016a). Sentence and Narrative Speech Production. In

Neurobiology of Language (pp. 751–762). Elsevier. [Link]

12-407794-2.00060-2

Wise, R. J. S., & Geranmayeh, F. (2016b). Sentence and Narrative Speech Production. In

Neurobiology of Language (pp. 751–762). Elsevier. [Link]

12-407794-2.00060-2

227
Woollams, A. M., Cooper-Pye, E., Hodges, J. R., & Patterson, K. (2008). Anomia: A doubly

typical signature of semantic dementia. Neuropsychologia, 46(10), 2503–2514.

[Link]

Wright, H. H., Koutsoftas, A. D., Capilouto, G. J., & Fergadiotis, G. (2014). Global

coherence in younger and older adults: Influence of cognitive processes and discourse

type. Aging, Neuropsychology, and Cognition, 21(2), 174–196.

[Link]

Wu, C.-L., Huang, S.-Y., Chen, P.-Z., & Chen, H.-C. (2020). A Systematic Review of

Creativity-Related Studies Applying the Remote Associates Test From 2000 to 2019.

Frontiers in Psychology, 11, 573432. [Link]

Wu, W., & Hoffman, P. (2022). Validated measures of semantic knowledge and semantic

control: Normative data from young and older adults for more than 300 semantic

judgements. Royal Society Open Science, 9(2), 211056.

[Link]

Wu, W., Lohani, S., Homan, T., Krieger-Redwood, K., & Hoffman, P. (2022). Healthy ageing

has divergent effects on verbal and non-verbal semantic cognition.

[Link]

Wu, W., Morales, M., Patel, T., Pickering, M. J., & Hoffman, P. (2022). Modulation of brain

activity by psycholinguistic information during naturalistic speech comprehension and

production. Cortex, 155, 287–306. [Link]

Xu, J., Moeller, S., Auerbach, E. J., Strupp, J., Smith, S. M., Feinberg, D. A., Yacoub, E., &

Uğurbil, K. (2013). Evaluation of slice accelerations using multiband echo planar

imaging at 3 T. Neuroimage, 83, 991–1001.

Yee, E., & Thompson-Schill, S. L. (2016). Putting concepts into context. Psychonomic

Bulletin & Review, 23(4), 1015–1027. [Link]

228
Yeo, B. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M.,

Roffman, J. L., Smoller, J. W., Zöllei, L., Polimeni, J. R., Fischl, B., Liu, H., &

Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by

intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165.

[Link]

Yeshurun, Y., Nguyen, M., & Hasson, U. (2021). The default mode network: Where the

idiosyncratic self meets the shared social world. Nature Reviews Neuroscience, 22(3),

181–192. [Link]

Yuspeh, R. L., & Vanderploeg, R. D. (2000). Spot-the-Word: A Measure for Estimating

Premorbid Intellectual Functioning.

Zabelina, D. L. (2018). Attention and Creativity. In R. E. Jung & O. Vartanian (Eds.), The

Cambridge Handbook of the Neuroscience of Creativity (1st ed., pp. 161–179).

Cambridge University Press. [Link]

Zabelina, D. L., Friedman, N. P., & Andrews-Hanna, J. (2019). Unity and diversity of

executive functions in creativity. Consciousness and Cognition, 68, 47–56.

[Link]

Zabelina, D. L., & Robinson, M. D. (2010). Creativity as flexible cognitive control.

Psychology of Aesthetics, Creativity, and the Arts, 4(3), 136–143.

[Link]

Zabelina, D. L., Robinson, M. D., Council, J. R., & Bresin, K. (2012). Patterning and

nonpatterning in creative cognition: Insights from performance in a random number

generation task. Psychology of Aesthetics, Creativity, and the Arts, 6, 137–145.

[Link]

Zedelius, C. M., Mills, C., & Schooler, J. W. (2019). Beyond subjective judgments:

Predicting evaluations of creative writing from computational linguistic features.

229
 Behavior Research Methods, 51(2), 879–894. [Link]

1137-1

Zhang, J., Zhuang, K., Sun, J., Liu, C., Fan, L., Wang, X., Gu, J., & Qiu, J. (2023). Retrieval

flexibility links to creativity: Evidence from computational linguistic measure.

Cerebral Cortex, 33(8), 4964–4976. [Link]

Zhang, M., Varga, D., Wang, X., Krieger-Redwood, K., Gouws, A., Smallwood, J., &

Jefferies, E. (2021). Knowing what you need to know in advance: The neural

processes underpinning flexible semantic retrieval of thematic and taxonomic

relations. NeuroImage, 224, 117405.

[Link]

Zhu, W., Chen, Q., Xia, L., Beaty, R. E., Yang, W., Tian, F., Sun, J., Cao, G., Zhang, Q.,

Chen, X., & Qiu, J. (2017). Common and distinct brain networks underlying verbal

and visual creativity. Human Brain Mapping, 38(4), 2094–2111.

[Link]

Zhuang, K., Yang, W., Li, Y., Zhang, J., Chen, Q., Meng, J., Wei, D., Sun, J., He, L., Mao, Y.,

Wang, X., Vatansever, D., & Qiu, J. (2021). Connectome-based evidence for creative

thinking as an emergent property of ordinary cognitive operations. NeuroImage, 227,

117632. [Link]

230
231
 Appendices

Appendix A: Supplemental Materials (Chapter 3)

Instructions for Creativity Ratings

Background

You are going to be presented with a series of original responses from a previous
study on Creativity. Each of the responses was produced by participants while they were
performed a classic creativity task called the ‘Alternate Uses Task’ (AUT). In the AUT,
participants are presented with a common, everyday object (e.g., ‘a rope’) and asked to
produce alternative uses for the object (e.g., ‘pile of rope as a pillow’).

What you will have to do

Your task is to read through each of the responses and rate each response on how
creative it is. You will be assigning each response a single score, using a scale of 1 (Not at
all Creative) to 5 (Highly Creative).

When you are scoring, you need to keep in mind three dimensions of creative ideas.
1. Uncommon: Creative ideas are uncommon; they will occur infrequently in the
sample of responses, and they will be unique. Note, just because a response is only given
once, it need not be judged as creative. For example, a random or inappropriate response
would be uncommon but not creative.
2. Remoteness: Creative ideas are remotely linked to everyday objects and ideas. For
example, creative uses for a brick are far from the common uses for a break. Responses
that stray from obvious ideas are more creative, and responses that are close to obvious
ideas are uncreative.
3. Clever: Creative ideas are often clever; they strike people as insightful, ironic,
humorous, fitting, or smart. Responses that are clever will tend to be creative responses.
Keep in mind that cleverness can compensate for other facets. For example, a common
use that is cleverly expressed could receive a high score.

1. To recap, when rating each item, think of the following dimensions of creativity:
uncommonness, remoteness, and cleverness. Strength in one dimension can balance
weakness in another.
2. You might come across some responses that have been repeated – try to mark all the
responses of one kind in the same way. (Unless the response is different in some way
from the other responses of the same kind, e.g., if it is more descriptive)
3. Give low scores to any responses that list the actual uses of the object. e.g., if the
object was a brick, and the response is ‘to make a wall’ or ‘to make a fireplace’, then you
would give these responses a score of 1 (Not at all Creative) as they are the actual uses of
the brick.
4. Give low scores to any responses that are incomplete, non-understandable, or
nonsensical. e.g., If the object was a brick, and the response was ‘to eat’, then you would
give this response a score of 1 (Not at all Creative) as you cannot eat a brick.
5. Give low scores to vague ideas. e.g., if the idea is under-developed and you cannot
form a full picture of the intended use with the information given.

232
Pearson’s Correlations for each age group

Figure S1: Pearson correlations between abilities. (a) Correlations in the older group. (b) Correlations
in the younger Note that Inhibition and Shifting used time-based measures so higher values indicate
poorer performance (unlike the other measures), RAT = Remote Associates Test; AUT = Alternate
Uses Test. m

* = p <.05; ** = p <.01; *** = p <.001.

233
Group-level models

Table S1: Analysis of effects of semantic and executive abilities on Alternate Uses Task (AUT)
Originality within the older and younger groups.
Effect Older Adults Younger Adults
B (se) CI p B (se) CI p
Semantic Selection -0.02 -0.07 – 0.02 0.307 0.07 0.02 – 0.11 0.003
Knowledge -0.01 -0.06 – 0.05 0.841 0.02 -0.03 – 0.06 0.477
Controlled Retrieval -0.01 -0.06 – 0.04 0.761 0.03 -0.02 – 0.08 0.216
AUT Elaboration 0.09 0.04 – 0.14 <.001 0.08 0.03 – 0.13 <.001
Updating 0.02 -0.04 – 0.08 0.548 0.02 -0.03 – 0.07 0.386
Inhibition 0.01 -0.04 – 0.06 0.707 -0.00 -0.05 – 0.04 0.894
Shifting 0.01 -0.04 – 0.06 0.636 0.04 -0.01 – 0.08 0.120
Fluid Intelligence 0.04 -0.01 – 0.10 0.112 -0.00 -0.05 – 0.05 0.962
Reasoning Speed 0.03 -0.03 – 0.09 0.324 0.01 -0.04 – 0.05 0.748

Table S2: Analysis of effects of semantic and executive abilities on Alternate Uses Task (AUT)
Subjective Ratings within the older and younger groups.
Effect Older Adults Younger Adults
B (se) CI p B (se) CI p
Semantic Selection -0.02 -0.05 – 0.01 0.283 0.03 0.01 – 0.06 0.005
Knowledge 0.01 -0.02 – 0.04 0.517 -0.01 -0.03 – 0.02 0.559
Controlled Retrieval -0.02 -0.05 – 0.01 0.185 0.04 0.01 – 0.06 0.004
AUT Elaboration 0.04 0.01 – 0.07 0.009 0.06 0.03 – 0.08 <.001
Updating 0.01 -0.03 – 0.05 0.501 -0.02 -0.05 – 0.00 0.107
Inhibition 0.04 0.00 – 0.07 0.032 0.01 -0.02 – 0.03 0.505
Shifting 0.01 -0.03 – 0.04 0.709 -0.01 -0.04 – 0.01 0.259
Fluid Intelligence -0.01 -0.04 – 0.03 0.602 -0.01 -0.03 – 0.02 0.517
Reasoning Speed 0.07 0.03 – 0.11 0.001 0.01 -0.01 – 0.04 0.317

234
Table S3: Analysis of effects of semantic and executive abilities on Alternate Uses Task (AUT)
Uniqueness within the older and younger groups.
Effect Older Adults Younger Adults
B (se) CI p B (se) CI p
Semantic Selection 0.00 -0.07 – 0.07 0.998 0.07 0.01 – 0.13 0.019
Knowledge -0.01 -0.08 – 0.06 0.871 0.04 -0.02 – 0.10 0.208
Controlled Retrieval 0.02 -0.05 – 0.09 0.520 -0.07 -0.13 – -0.01 0.031
AUT Elaboration 0.04 -0.03 – 0.11 0.244 0.03 -0.03 – 0.09 0.331
Updating 0.02 -0.06 – 0.11 0.619 0.05 -0.02 – 0.12 0.145
Inhibition -0.04 -0.11 – 0.04 0.328 -0.07 -0.13 – -0.01 0.033
Shifting -0.02 -0.10 – 0.05 0.516 0.02 -0.04 – 0.09 0.418
Fluid Intelligence 0.00 -0.07 – 0.08 0.959 -0.03 -0.09 – 0.03 0.364
Reasoning Speed 0.04 -0.05 – 0.12 0.400 -0.04 -0.10 – 0.02 0.158

Table S4: Analysis of effects of semantic and executive abilities on Alternate Uses Task (AUT)
Fluency within the older and younger groups.
Effect Older Adults Younger Adults
B (se) CI p B (se) CI p
Semantic Selection 0.52 -0.48 – 1.52 0.301 0.32 -0.52 – 1.16 0.449
Knowledge -0.40 -1.47 – 0.67 0.454 0.03 -0.83 – 0.89 0.944
Controlled Retrieval 0.25 -0.80 – 1.31 0.633 0.41 -0.50 – 1.32 0.365
AUT Elaboration -0.28 -1.29 – 0.72 0.573 -0.09 -0.98 – 0.80 0.845
Updating 0.26 -1.01 – 1.54 0.680 0.34 -0.63 – 1.30 0.483
Inhibition -0.39 -1.51 – 0.73 0.484 -0.20 -1.07 – 0.68 0.654
Shifting -0.29 -1.38 – 0.80 0.594 -0.27 -1.14 – 0.61 0.547
Fluid Intelligence 1.06 -0.08 – 2.21 0.069 -0.27 -1.16 – 0.61 0.536
Reasoning Speed -0.09 -1.41 – 1.22 0.889 0.00 -0.87 – 0.88 0.992

235
 Appendix B: Supplemental Materials (Chapter 4)

Generation Task: Instructions and Materials

In this section, you will be presented with three target words. Your task is to generate as
many solution words as possible, that could be related to all three cues. In the given response
box, please enter all the words you consider while searching for a solution. You will be given
two minutes per question. Don't stop even if you think you've got the answer, keep going for
the whole two minutes!

Table S5: Controlled Retrieval Task: List of Trials

No. Cue 1 Cue 2 Cue 3

1 irk butterflies audacity

2 slice pool post
3 molten film treat
4 drift expose ruin
5 wind puppet scalpel
6 symptom protect indulgence
7 community component station
8 cactus column foundation
9 clump weight public
10 insult shear interrupt

Table S6: Controlled Retrieval Task: List of Trials

No. Cue 1 Cue 2 Cue 3 Target % solved

1 Time Hair Stretch long 16.22

2 Salt Deep Foam sea 72.07

3 Pure Blue Fall water 25.23

4 Blade Witted Weary dull 4.50

5 Silk Cream Even smooth 32.43

6 Thread Pine Pain needle 37.84

7 Cloth Sad Out sack 4.50

8 Goat Pass Green mountain 8.11

9 Poke Go Molasses slow 4.50

10 Stone Door Street house 23.42

236
Evaluation Task: Instructions and Materials

In this section, you will be presented with three cue words at the top of the page, all of
which are semantically related to each other in some way. Below the cues, you will be given
three possible solution words. Please click on the most appropriate solution word that relates
to all three cues. You will have 30 seconds to respond to each trial.
Table S7: Evaluation Task: List of Trials

No. Cue 1 Cue 2 Cue 3 Target Distractor Unrelated Pilot % Study %

solved solved
1 dull monitor idle drone screen favour 0.50 36.94
2 order brief sign tell method fever 0.53 49.55
3 quarrel command oath word leader basket 0.63 49.55
4 passage secure volume content journey insect 0.63 65.77
5 lower kid frivolous minor degrade airport 0.77 79.28
6 urge marbles spring mind ball photo 0.60 38.74
7 wrap hamper contrast foil picnic error 0.67 54.05
8 strain line shoot streak film cookie 0.60 36.94
9 bite smart firm crisp company orbit 0.77 57.66
10 ability steer command head military curtain 0.67 36.94
11 bat dawn move strike relocate dairy 0.70 63.96
12 plight bottle steep pickle rapid charter 0.73 35.14
13 capture gather will draw desire braid 0.47 22.52
14 predict alert faint fear weather album 0.53 60.36
15 state shell map frame nation humour 0.47 20.72
16 pitch dampen air tone field scandal 0.53 38.74
17 chain press firm iron franchise harbour 0.70 63.06
18 coil vary complex twist electric duty 0.70 65.77
19 quality custom definitive standard business drama 0.83 56.76
20 pick model bloom prime version bishop 0.63 45.05

237
Appendix C: Supplemental Materials (Chapter 5)
Table S8: Demographic summary and list of prompts used across datasets.

Code Participants Datasets Prompt

What would it have been like to live in the Middle Ages?
Ages 127 MD, EL, NC
What would it be like to live in Antarctica?
Antarctica 126 MD, EL, NC
What do the police do when a crime has been committed?
Crime 125 MD, EL, NC
What sort of things do you have to do to look after a dog?
Dog 127 MD, EL, NC
Do you think it's a good idea to send people to live on Mars?
Mars 127 MD, EL, NC
What do people usually do when getting ready for work in the morning?
Morning 125 MD, EL, NC
Describe a typical visit to a restaurant.
Restaurant 126 MD, EL, NC
Describe why someone might travel to Scotland on holiday
Scotland 127 MD, EL, NC
Which is your favourite season and why?
Season 125 MD, EL, NC
What happens when a storm is forecast in the UK?
Storm 125 MD, EL, NC
Describe the steps you would need to take if going somewhere by train.
Train 126 MD, EL, NC
What are the advantages and disadvantages of going to university?
University 128 MD, EL, NC
What do people usually do on Christmas Day in the UK?
Christmas 72 EL, NC
What sort of things does the Queen do on a typical day?
Queen 75 EL, NC
Election 15 NC What happens during a general election in the UK?

238
 Describe how you would make a cup of tea or coffee
Beverage 40 NC, SL
What sort of things usually happen at a wedding?
Wedding 40 NC, SL
Why are some people concerned about climate change?
Climate 40 NC, SL
Do you think the internet has improved people's lives?
Internet 39 NC, SL
How would you prepare to go on holiday?
Holiday 39 NC, SL
Why is it important to have a balanced diet?
Diet 25 SL
Describe the steps you’d take to order-in food.
Food 25 SL
Describe a typical visit to a grocery store.
Grocery 25 SL
What would you recommend doing during a job interview?
Interview 25 SL
What do people usually do on New Year’s Eve in the UK?
New Year 25 SL
What sorts of things do people do to cope with stress?
Stress 25 SL
What sort of things does a teacher do when at work?
Teacher 24 SL

239
240