that’s frowned upon - lot publicationsthat’s frowned upon using facial emg to track evaluation...
TRANSCRIPT
That’s frowned upon
Using facial EMG to track evaluation and simulation
during affective language processing
Published by
LOT phone: +31 30 253 6111
Trans 10
3512 JK Utrecht e‐mail: [email protected]
The Netherlands http://www.lotschool.nl Cover illustration: Balloërveld, Drenthe.
ISBN: 978‐94‐6093‐253‐3
NUR: 616
Copyright © 2017: Björn ‘t Hart. All rights reserved.
That’s frowned upon
Using facial EMG to track evaluation and simulation during affective
language processing
Afkeurende blikken
Het volgen van simulatie en evaluatie gedurende affectieve
taalverwerking met behulp van gezichts‐EMG
(met een samenvatting in het Nederlands)
Proefschrift
ter verkrijging van de graad van doctor aan de Universiteit Utrecht
op gezag van de rector magnificus, prof. dr. G.J. van der Zwaan, ingevolge
het besluit van het college voor promoties
in het openbaar te verdedigen op
vrijdag 29 september 2017
des middags te 2.30 uur
door
Björn ’t Hart
geboren 9 oktober 1984
te Assen
Promotor: Prof. dr. J.J.A. van Berkum
Copromotor: Dr. M.E. Struiksma
The research reported here was part of the project ‘Moving the
language user – Affect and perspective in discourse processing’
supported by the Netherlands Organisation for Scientific Research
under project number 277‐89‐001.
in memory of my father
and his parents
Table of Contents
Acknowledgements ................................................................................................... ix
1. Introduction ...................................................................................................... 1
1.1 Grounded cognition and language comprehension .......................................... 3
1.1.1 Simulating perception language ........................................................... 4
1.1.2 Simulating action language ................................................................... 6
1.1.3 Simulating emotion language ............................................................... 6
1.1.4 Simulation and comprehension ............................................................ 7
1.2 Theories of emotion .......................................................................................... 8
1.2.1 Classifying emotion ............................................................................. 11
1.2.2 Measuring emotion using facial electromyography ........................... 12
3.1. Language comprehension and narrative ........................................................ 14
1.3.1 Defining narrative ............................................................................... 14
1.3.2 Narrative as mimicry of experience .................................................... 16
1.4 A language processing framework for facial EMG activity .............................. 19
1.4.1 Simulation‐only model ........................................................................ 20
1.4.2 Evaluation‐blocks‐simulation model .................................................. 21
1.4.3 Multiple‐drivers model ....................................................................... 22
2. Study 1 ............................................................................................................ 23
2.1. Introduction .................................................................................................... 23
2.2. Method ........................................................................................................... 30
2.2.1 Participants and Stimulus Material ..................................................... 30
2.2.2 Procedure and Data Acquisition ......................................................... 31
2.2.3 Data Preparation and Analysis ............................................................ 31
2.3. Results ............................................................................................................. 34
2.3.1 Character Morality Manipulation ....................................................... 34
2.3.2 Critical Event Manipulation ................................................................ 35
2.3.3 Individual Differences ......................................................................... 38
2.4. Discussion ....................................................................................................... 39
3. Study 2 ............................................................................................................ 47
3.1 Introduction .................................................................................................... 47
3.2 Method ........................................................................................................... 55
3.2.1 Participants and Stimulus Materials ................................................... 55
3.2.2 Procedure and Data Acquisition ......................................................... 55
3.2.3 Data Preparation and Analysis ............................................................ 56
3.3 Results ............................................................................................................. 58
3.3.1 Character Morality .............................................................................. 58
3.3.2 Affective State Adjective ..................................................................... 60
3.3.3 Post‐hoc Analysis of Neutral Segment. ............................................... 61
3.3.4 Affect Reason ...................................................................................... 62
3.4 Discussion ....................................................................................................... 64
3.4.1 Frowning upon Immoral Behaviour .................................................... 65
3.4.2 Processing Character Affect ................................................................ 65
3.4.3 Processing Reasons for Character Affect ............................................ 67
3.4.4 Waiting to Update the Situation Model?............................................ 68
3.4.5 Open Questions .................................................................................. 68
3.4.6 Conclusion .......................................................................................... 69
4. Study 3 ............................................................................................................ 71
4.1 Introduction .................................................................................................... 71
4.2 Method ........................................................................................................... 76
4.2.1 Participants and Stimulus Materials ................................................... 76
4.2.2 Procedure and Data Acquisition ......................................................... 77
4.2.3 Data Preparation and Analysis ............................................................ 79
4.3 Results ............................................................................................................. 80
4.3.1 Character Manipulation ...................................................................... 80
4.3.2 Affective State Adjective ..................................................................... 83
4.3.3 Affect Reason Segment ....................................................................... 86
4.4 Discussion ....................................................................................................... 90
4.4.1 Morality vs. Minimal Group Manipulation ......................................... 90
4.4.2 Processing Character Affect ................................................................ 91
4.4.3 Processing Reasons for Character Affect ............................................ 92
4.4.4 General Discussion .............................................................................. 94
5. General Discussion .......................................................................................... 97
5.1 Facial EMG during online affective language comprehension ........................ 97
5.1.1 Simulation‐only model ........................................................................ 98
5.1.2 Evaluation‐blocks‐simulation model ................................................ 101
5.1.3 Multiple‐drivers model ..................................................................... 103
5.2 Complex evaluation explanation ................................................................... 107
5.3 Simulation: lexical concepts or situation models? ........................................ 108
5.4 Simulation and language comprehension ..................................................... 109
5.5 Character manipulation ................................................................................ 112
5.6 Concluding Remarks ...................................................................................... 113
References .............................................................................................................. 115
Supplementary Information A ................................................................................ 133
Supplementary Information B ................................................................................ 172
Supplementary Information C ................................................................................ 176
Supplementary Information D ................................................................................ 199
Supplementary Information E ................................................................................ 205
Samenvatting in het Nederlands ............................................................................ 212
Curriculum Vitae ..................................................................................................... 219
ix
Acknowledgements
The years spent working on this book have been some of the happiest in my life to
date. They were also some of the most challenging and frustrating and there were
many times where it seemed highly unlikely to me that I would ever complete the
work. To be perfectly honest, the fact that this book finally came about at all is
thanks to a very large number of people. In the next few pages I will attempt to
give them all their dues. Please accept my sincerest apologies in advance if I’ve left
anyone out.
The first person I would like to thank is my promotor, Jos van Berkum. To
start with, perhaps somewhat belatedly, thank you for taking a chance on a
somewhat unlikely applicant. I will always treasure this time because it allowed me
to become the person I had long since forgotten I could or wanted to be. You
pushed me and supported me, complimenting and criticising as needed. You
guided me in my academic life, but also listened and reserved space for the various
things, good and bad, that cropped up in my private life. Thank you for all of that
and much more.
I am also intensely grateful to my wonderful co‐promotor Marijn
Struiksma. Without your daily guidance and supervision, this book would not exist.
Thank you for being patient with my endless impromptu visits to your office and for
always taking the time to listen to me and think along with me. Thank you also for
boosting my confidence at critical times and reminding me of all the progress I had
already made. I am honoured to have been the first PhD student you supervised
and I am certain I will not be the last. They will be very lucky indeed!
I am also incredibly thankful to the other amazing members of the VICI
team. Anne van Leeuwen, thank you for making me feel welcome from day one.
We had lunch the day I interviewed for this position and I was so nervous, but you
made me feel like I could totally be a PhD student too. We shared so much of our
journey and I can’t wait to celebrate your defence in December! Hannah de
x
Mulder, I want to thank you for many things, but I will pick three. Thank you for
allowing me to benefit from your sharp insights, for sharing your vulgar sense of
humour, and for being my drinking buddy on more occasions than were probably
wise. Iris Mulders, thank you for helping to make my lab research possible and for
being my friend. You are a beautiful person and having your friendship and respect
makes me feel proud.
The next five people have also been instrumental in this book actually
coming about. Hans Hoeken, Eggo Mueller, Frank Wijnen, Roel Willems, and Rolf
Zwaan, thank you all for agreeing to be on my reading committee and dedicating
your time and energy to reading the manuscript.
Next, I would like to express my gratitude to the various people who have shared
my office over the years. Kiki Kushartanti, thank you for helping me get settled in
and figure out how things worked. Anna Volkova, thank you for your grumpy,
almost‐finished PhD wisdom and for shouting at Russia on my left for a couple of
months. I am glad we have continued our friendship in the last few years. Jingwei
Zhang, thank you for your calm and friendly presence. Mulusew Wondem, thank
you for carving out time to talk to me when you were so very busy. Luying Hou,
thank you for always being happy to see me and have a chat. Dominique Blok,
thank you for working next to me, at the office and at the Hallen. Also, thanks for
sharing the secret of the restorative powers of lying flat on your back for fifteen
minutes in the middle of the day.
Two of my office mates require special attention. We shared office 2.24b
for the longest time and together we made it the healthiest office at Trans. Jolien
Scholten, I am going to miss our daily chats and doing stretches with you. Anja
Goldschmidt, I am going to miss our wide‐ranging discussions and even your
incredibly detailed explanations of things that annoy you. Thank you both for being
such wonderful daily companions during this amazing time, we shared so much of
this crazy journey and I will miss you!
xi
The UiL OTS has been a wonderful bubble of special space and time filled with a
great many clever, funny, and interesting people. So many in fact, that I can’t
possibly thank them all properly, but it was the presence of all of you that made
the last four‐plus years so magical. The following list will fail to do justice to each
individual’s awesomeness, but it can’t be helped, sorry!
In no particular order; thank you Marjolein van Egmond, Desiree Capel,
Assaf Toledo, Brigitta Keij, Liquan Liu, Tom Lentz, Hayo Terband, Carolien van den
Hazelkamp, Myrthe Bergstra, Andrea Santana Covarrubias, Suzanne Kleijn,
Gerdineke van Silfhout, Rogier Kraf, Fang Li, Franca Wessling, Hans Rutger Bosker,
Renske Bouwer, Monica Koster, Liv Person, Marko Hladnik, Louise Nell, Sandrien
van Ommen, Marloes Herrijgers, Yipu Wei, Shuangshuang Hu, Mengru Han, Jet
Hoek, Jorik van Engeland. I have shared PhD struggles, drinks, and laughs with all
of you, I have attended your talks and you have attended mine. I am grateful. I also
owe a special thank you to Chris van Run, Martijn van der Klis, and Maarten
Duijndam. Without your help in the lab I would have never been able to the
research I did, thank you so much! Heartfelt thanks also go to Anton van Boxtel, my
advisor on all things EMG, and Huub van den Bergh, who patiently explained mixed
models to me again and again. There have also been various student assistants
over the years that have helped with my research; Sanne Riemsma, Lea ter Meulen,
Myrthe Bergstra, Jorik Geutjes, Ella Bosh, en Eletta Daemen; thank you all so much
for your help making the practical side of my experiments possible.
Alexia Guerra Riviera, you are a such a wonderful, warm, intelligent and
fun woman. The chats we had helped me through the middle part of my project
and I look forward to visiting you in Chile one day. Zenghui Liu, my running buddy!
Thank you for your kindness and humour, thank you for the tea, talking mixed
models with me, and thank you for complimenting my hair. Marko Simonovic,
nothing gets the brain juices flowing like a conversation with you. Thanks for
baffling me and for making me laugh. Mirjam Hachem, you wonderful, crazy,
intense woman! I’d like to thank you properly, but “I don’t know how!” Maartje
xii
Schulpen, thank you for telling me you practiced saying my name in the shower and
telling me you liked my voice. I’m so glad we met. Stavroula Alexandropolou, you
always seemed somewhat mysterious to me, but you are a wonderful warm person
and I want to thank you for our many chats, especially those where you helped me
silence the self‐doubt.
To those of you who were a part of Kibbutz, that wonderful food
collective, thank you for feeding me (no really, thank you!) and providing a warm
and wonderful atmosphere. You are a huge part of why these last couple of years
have been so special and in your company, I was able to be who I wanted to be.
You know who you are.
Heidi Klockmann, there are so many things I want to thank you for. Thank
you for the many coffees where we aired our mutual insecurities and doubts.
Thank you for discussing life matters with me, I feel like we have a lot in common in
many areas of life and talking to you has helped in ways I can’t fully explain. I am
grateful for our friendship, you are one of the few people I know who are able to so
unabashedly be themselves; don’t ever let anyone tell you to be any different,
you’re amazing and I am so proud to have been your paranymph!
And then we come to my paranymphs: Marta Castella and Anna Sara Hexeberg
Romøren. You two…what can I say?
Marta, we met at a very turbulent time for you; the winter of your discontent.
Nonetheless, you made a space for me in your life, thank you for that. You
occupied the desk next to mine for a while and our friendship grew slowly, bonding
over cooking and baking. I especially enjoyed that one summer when you were
funny. Some of the happiest times in my life have been spent in the kitchen with
you or sharing music over a glass of wine. Thank you for teaching me how to bake
decent bread, for helping me shop, for listening to me, and for understanding that I
xiii
am not really all that calm and collected at all. Let me just say you are one of the
most beautiful individuals I have ever met.
Anna Sara, our morning coffees started during your time at Trans and at times they
were the only reason I actually got up and went to work. You are gorgeous, inside
and out, especially when you blush. I do believe I have always had a bit of a crush
on you. Thank you for your sense of humour, thank you for sharing your troubles
with me and listening to mine. Thank you also for believing me when I said I was
going to be your friend no matter what when your life was in upheaval. I meant it
then and I mean it now. I hope we can continue our morning coffees, whether in
person or over WhatsApp as we commute to work in different countries.
Thank you both for agreeing to be my paranymphs. I could not imagine my defence
any other way than with the two of you behind me. I love you both like sisters and
you have a place in my heart forever.
I would also like to thank some of my oldest friends, those from the ‘before times’.
Before I disappeared into the gaping maw of academia. Nick Degens, you and I
have been friends ever since I can remember and your presence in my life
throughout all of its phases provides a red thread that grounds me. You and I share
so much in common, both as people and in terms of history, that you are an
integral part of who I am. I treasure our friendship immensely.
Karlien Haak, we have also been friends for such a long time. During the
period when we were neighbours, our weekly dinners and movie/TV marathons
saved my life. That sounds like a dramatic exaggeration, but it is honestly how it
feels looking back. You are an amazingly brave, funny, and insightful woman, you
hold a special place in my heart and I feel like a part of me appears when you’re
around that never surfaces otherwise. Thank you for everything.
xiv
Gees Maathuis, voor jou even een klein uitstapje naar het Nederlands
zodat je goed begrijpt hoeveel je voor me betekent. Onze vriendschap gaat
ondertussen meer dan tien jaar terug en in die tijd heb je me laten lachen, heb je
me te eten gegeven, heb je jouw sores met mij gedeeld, en heb je die van mij
aangehoord. Je bent één van de meest authentieke en bijzondere mensen die ik
ken en het feit dat jij me een plekje in jouw hart hebt gegeven maakt me meer dan
trots. Bedankt voor alles, en in het bijzonder ook voor het verzorgen van een plek
waar ik lekker plat kan praten!
Maria Barkman, our friendship began in high school, during art class. We
spent more time drinking coffee or hot chocolate and smoking outside than
drawing or painting. Although not among your biggest talents, that kabuki‐duck
tableau you drew was phenomenal, pure genius. You were there when my father
passed away, letting me sleep in your bed while I drank my way through grief. I
can’t thank you enough for that and for your friendship through the years. Your
beautiful daughter, Abby, is also the first child that made it into my heart, I am
looking forward to our continued friendship and getting to watch Abby grow up.
Marc Lansdown, your steadfast confidence in my abilities has helped me
gain confidence over the ten‐plus years we’ve known each other. Especially your
often‐restated assertion that I am your most intelligent friend! I love our catch‐ups
over burgers where you tell me stories that are way too long (but you also don’t
mind me interrupting you and telling you so). Did you know some of my other
friends know you as Burger Boy? You have been there for so many of the struggles
in my life and I am grateful for your friendship and support.
Margriet Hesseling, we became friends all those years ago in Groningen
without me even noticing. Thank you for staying in touch. Your optimism and
perseverance have been an inspiration to me. I also would be very grateful if you
could please make me some of that pecan pie again sometime soon!
xv
Alie Drenthe, Thank you for your friendship, for the food, the wine, the
cryptograms, the refuge, and for your daughter. I miss you. I hear your voice in my
head at least once a week: “Jamie, alles is goed.”
I dedicate this book to the memory of my father, Roel ‘t Hart, grandmother, Cor ‘t
Hart, and grandfather, Wim ‘t Hart. I miss them and I can’t express what needs to
be said here; nothing I have done would have ever been possible without them.
The same goes for my wonderful mother, Pietie Boonstra. Dankjewel moeders,
voor je stille steun, voor je kritiek, voor je liefde. Dankjewel voor de uren die je aan
mijn bed hebt gezeten toen ik pijn had, voor de angsten die je hebt uitgestaan toen
ik het echt niet meer wist, voor de vrijheid die je me hebt gegeven om fouten te
maken, voor de fouten die je hebt toegegeven, en voor alle talloze keren dat je me
hebt geholpen om mijn fouten weer recht te zetten. Ik hou van je.
Carlo ‘t Hart, broeder, ondanks dat we elkaar weinig zien ben je toch mijn grote
broer en ben je altijd in mijn gedachten. Hetzelfde geldt voor jou, Sylvia Tepper,
jouw vriendschap was jarenlang het belangrijkste in mijn leven en ook jij bent altijd
in mijn hart. Ik ben blij dat jullie elkaar hebben gevonden en wens jullie alle goeds.
The last person I want to thank is the love of my life, Thomas William Atkins, my
sweet Tom (a.k.a. Wommus). I would never have even applied for a PhD position if
you hadn’t encouraged me. You believe in me so much and I am truly grateful for
your faith, patience, and support. I love all of the many silly love songs you have
written for me. Nobody knows me like you do. You saw things in me all those years
ago that I am only just discovering myself. I would never have achieved as much,
academically or personally, if I had not had you in my life. You made a serious boy
more fun, you made a pessimist more optimistic, and you showed me a kind of love
that I hoped, but didn’t believe, existed. You are a beautiful man with a spontaneity
xvi
that almost everybody I know lost along the way to adulthood. You are clever, wise
and silly. You buy too much crockery and glassware, but you have created a home
for us that makes me feel warm inside. You are skilled and adept in all the areas of
life that count and where I have always felt inadequate. I know I sometimes
criticise you for things that don’t really matter. You are actually a much better
person than I am. And yet, when you look at me I feel like the most amazing person
ever. Thank you for that. I am so excited for our future. I love you so much.
xvii
1
1. Introduction
“I knew I had fallen in love with Lolita forever; but I also knew she would not be
forever Lolita.”
Vladimir Nabokov, Lolita
Stories can make us intimate witnesses to the emotional life of the characters that
populate them. More than that, stories can transport us into another world where
we vicariously experience the overwhelming and hopeless love felt by this
character from Nabokov’s Lolita. Put simply, stories provide the opportunity to feel
what it is like to be that character and fall so madly and precipitously in love.
Theories of language comprehension have to account for how we manage
to represent, as in our example, a concept such as love during real‐time language
processing with no more to go on than a collection of letters that have no intrinsic
connection to the emotion they refer to. According to one of the dominant
theories in the field today, grounded cognition, our understanding of the meaning
of the word ‘love’ involves simulation, a kind of mental re‐enactment of our
previous experiences of love. Before going into more detail, the core notion of
simulation is that, according to grounded accounts of language comprehension, our
vicarious experience is central to how we understand what the story is telling us.
However, the idea of vicarious experience as the basis for our
representation of emotion concepts during real‐time language comprehension is
not without complication. The notion of vicarious experience must imply that
language comprehension involves more than a cold, analytical process of
understanding; we do more than simply dispassionately comprehend language.
This is especially true when the language in question describes the emotions of
characters, real or imagined, in a story. We evaluate what we read and we respond
emotionally to it. Not infrequently, the emotions characters are described to have
2
will be at odds with our own (emotional) evaluation of those emotions. It is this
conflict between the meaning of language referring to emotion and our
emotionally coloured evaluation of it that is central to this dissertation.
The quote from Lolita is a case in point. The character whose all‐
consuming and frustrated feelings of love are the subject of the quote at the start
of this chapter is Humbert Humbert, a literature professor of middle years. There is
nothing necessarily problematic about a man falling in love, even if he does happen
to be a literary scholar. The object of his affection, however, is all the more
controversial. In his novel Lolita, Vladimir Nabokov describes Humbert Humbert’s
love for 12‐year‐old Dolores (Lolita) Haze. While beautifully written and widely
recognised as a classic 20th century work of literature, the book has been thrust
aside in disgust by more than one reader, including yours truly. Any simulation or
vicarious experience of ‘love’ in the quote at the start of this chapter is clearly at
odds with our own emotionally‐charged evaluation of this particular love.
The research in this thesis investigates the conflict between language‐
driven simulation of emotion concepts and our own emotional evaluation of those
concepts. It does so making use of narratives, because, as the example illustrates,
this kind of complexity of affective meaning arises frequently and naturally in the
stories that we read. Special attention is given to how this complexity impacts a
grounded account of language comprehension. Before turning to the reports of
three experiments designed to investigate how this affective complexity plays out
in online language comprehension, the introduction will provide necessary
background information on theories from three key domains.
Section 1.1 introduces the theory of grounded cognition and discuss its
relation to previous theories of cognition, including examples of evidence of
simulation in language comprehension. Section 1.2 discusses important theories of
emotion, followed by a working definition of emotion and the way this thesis
measures emotion. As in the example, the language comprehension research in
this thesis makes use of narrative stimuli. Section 1.3 summarises central theories
3
of narrative and provides a definition as well as a rationale for using it as a medium
in this research. Section 1.4 combines the theoretical and methodological
information from the preceding sections into a model for affective language
comprehension, including language‐driven simulation and emotional evaluation as
potential drivers for emotional responding during online language processing. This
model generates testable predictions, three of which are spelled out.
1.1 Grounded cognition and language comprehension
Until the latter part of the 20th century, standard theories of cognition held that our
knowledge of concepts was stored as amodal symbols in semantic memory. To pick
up the example of Humbert Humbert being in love, according to such theories an
amodal symbol‐like concept of love is stored somewhere in your brain and
arbitrarily linked to a sign (i.e., a particular word). This arbitrary link allows you to
read the word ‘love’ and understand what it is that the character to whom it
pertains is feeling. A fundamental problem with these theories has always been
how such symbols relate to perception, action, and experience; the so‐called
grounding problem (e.g., Harnad, 1990). Grounded cognition refers to a collection
of related theories of human cognition that specifically deal with this grounding
problem. Proponents of grounded cognition reject models of cognition where
conceptual knowledge is represented in amodal symbols. Instead, the basic tenet
of grounded cognition is a link between cognition and modality‐specific systems,
the body, the physical environment and the social environment (Barsalou, 2008).
According to grounded theories of cognition, real‐time conceptual
processing relies on modality specific systems. That is to say, it makes use of the
very same systems that support perception, action and internal states in the first
place. Anderson (2010) referred to this as neural reuse and Barsalou (2016) links
the notion of neural reuse to what has commonly been referred to as simulation
(Barsalou, 2003b; Barasalou, 2008). Concretely, according to this theory you
4
understand what the word ‘love’ refers to by activating traces of previous
experience in the same neural systems that were engaged when you felt and dealt
with love1.
The next three sections provide illustrations of evidence supporting the
existence of simulation in modality‐specific brain systems during language
comprehension. This evidence is organised according to the three kinds of language
mentioned in the definition of simulation by Barsalou (2008): perception language,
action language, emotion language.
1.1.1 Simulating perception language
Perception language is language that refers to aspects of reality that we perceive
with our senses, for instance sight, hearing, smell or taste. One simple, yet elegant
study that demonstrates that mental simulation is part of perceptual language
comprehension made use of a sentence‐picture verification task. Zwaan, Stanfield
& Yaxley (2002) found that after participants read a sentence such as “the ranger
saw the eagle in the sky,” they were faster to verify a picture of an eagle with wings
outstretched than one of an eagle with wings folded back. This shows that their
representation of the sentence contains implicit information about perceptual
features, such as shape. Importantly, Zwaan & Pecher (2012) showed this effect of
mental simulation to be robust in replication studies. Another, more recent study
(de Koning, Wassenburg, Bos, & van der Schoot, 2016) showed that in addition to
shape, implied object size is also mentally simulated. They found that participants
responded faster to a picture of a large statue after reading the sentence “the man
saw the statue in the garden” than when they had just read “the man saw the
1 I would argue that even without specific experience with a particular concept, combining traces of other, related experiences will nonetheless allow you to represent new concepts. This also relates to a grounded theory of language acquisition rather than language processing.
5
statue in the windowsill.” Interestingly, they found that while children between the
ages of 8 and 13 were overall slower to respond in this task, the match effect was
the same for both children and adults.
Montoro et al. (Montoro, Contreras, & Elosúa, 2015) also found evidence
for cross‐modal metaphorical mapping of emotion‐related language to the visual
perceptual domain; during active listening, participants responded faster to
positive than negative words when they were mapped ‘up’ on the visual axis rather
than ‘down’. This suggests that positive and negative valence are metaphorically
grounded in vertical space. Another study found that cross‐modal switching incurs
a processing delay (Pecher, Zeelenberg, & Barsalou, 2003). For instance, verifying a
property in one perceptual domain (e.g., blender‐loud) slows down the subsequent
verification of a property in another domain (e.g., cranberries‐tart) but not in the
same domain (e.g., leaves‐rustling). This switching cost is taken to support the idea
that modality‐specific simulation plays a part in language processing.
In addition to behavioural evidence, neuroimaging evidence also supports
the notion of modality‐specific simulation. For instance, Simmons et al. (2007)
found that a verbal object‐property verification task related to colour activated the
same areas of the brain involved in colour perception, while a verification task
related to movement did not. That is to say, judging whether the word ‘purple’
describes a property of ‘eggplant’ activates those same areas in the brain that are
involved in the perception of the colour purple. Such evidence of activation in
specific sensory systems while processing linguistic stimuli related to perception is
not restricted to the visual domain. There are also studies that show the same
effect for the processing of auditory and olfactory language (e.g., González, et al.,
2006; Kiefer, Sim, Herrnberger, Grothe, & Hoenig, 2008).
6
1.1.2 Simulating action language
Action language refers to motor activity, for instance walking and running or
pushing and pulling. One line of evidence has used functional neuroimaging
techniques to investigate how the processing of words related to actions executed
with specific effectors influences activity in motor systems. Several fMRI studies
(e.g., Aziz‐Zadeh, Wilson, Rizzolatti, & Iacobini, 2006; Hauk, Johnsrude, &
Pulvermuller, 2004) demonstrated that when participants read about an action, the
motor system is also active. Moreover, compared to a control condition, language
related to specific actions executed with, for example, the legs or arms evokes
activity specifically in those areas of the brain that are active when those effectors
are actually used.
In addition to neuroimaging evidence, behavioural evidence for action
simulation also abounds. The so‐called action sentence compatibility effect (ACE)
effect refers to the finding that responding to a linguistic stimulus with an action
that mimics the action described in that stimulus is facilitated (Glenberg & Kaschak,
Grounding language in action, 2002). More concretely, participants respond more
quickly when verifying the grammaticality of a sentence such as “The man closed
the drawer,” if the correct response is to push a lever away from them than when
the correct response involved pulling the lever toward the body. This result has
been successfully replicated repeatedly (e.g. Borghi & Riggio, 2009; Borreggine &
Kaschak, 2006; Zwaan & Taylor, 2006).
1.1.3 Simulating emotion language
Although the majority of research on grounded language comprehension initially
focussed on perceptual and action language, emotion language is also explicitly
included in the grounded enterprise (e.g., Vigliocco, Meteyard, Andrews, & Kousta,
7
2009). Emotion language refers to linguistic stimuli that either describe or evoke
emotionally salient internal states. This includes language describing short‐lived
emotions (e.g., anger, fear, happiness), but also mood (e.g., gloomy, sombre, calm),
action language related to emotionally salient internal states (smile, frown,
tremble), and language that does not refer to emotion directly, but is commonly
associated with an emotional response (e.g., party, birthday, vomit). Rather than
the seemingly narrow scope of emotion language, this thesis will refer to the
linguistic stimuli listed above as affective language more broadly.
There is ample evidence that suggests that affective language
comprehension involves simulation. One particularly fruitful line of evidence stems
from research using facial electromyography (EMG). What these studies show is
that reading language describing affective states is accompanied by facial muscle
activity associated with those same states. This effect has been demonstrated for
affectively salient words, including adjectives, verbs and nouns (e.g., Foroni &
Semin, 2009; Künecke, Sommer, Schacht, & Palazova, 2015), noun verb pairs (e.g.,
Fino, Menegatti, Avenanti, & Rubini, 2016), and complete sentences (e.g.,
Glenberg, Webster, Mouilso, Havas, & Lindeman, 2009; Havas, Glenberg, Gutowski,
Lucarelli, & Davidson, 2010). In these cases facial EMG activity is interpreted as a
downstream effect of the simulation of described or implied affect in parts of the
brain that are responsible for controlling the relevant facial muscles involved in the
expression of that affect. Section 1.3 provides a more in‐depth description of the
muscles involved in these facial EMG studies as well as detailed methodological
information.
1.1.4 Simulation and comprehension
The preceding three sections have provided a brief summary of evidence of
modality specific simulation in relation to language processing. While there is
ample evidence that suggests such simulation exists, whether simulation is a
8
necessary component of comprehension is another matter; simulation effects may
simply be epiphenomenal to comprehension. The question of whether simulation
contributes causally to language comprehension is the subject of an ongoing
theoretical and empirical debate (for recent discussions see Mahon & Hickok, 2016;
Leshinskaya & Caramazza, 2016; Barsalou, 2016; Zwaan, 2016).
The focus in the studies reported in this dissertation is on an investigation
of simulation and how it interacts with evaluation in the case of affective language.
This dissertation does not deal with the matter of a causal role for simulation in
language comprehension directly. We assume that simulation is attendant on
language processing and do not address the precise nature of the role it plays in
comprehension. However, a theory of grounded language processing that includes
simulation, especially as a causal component, will have to account for the type of
conflict between simulation and evaluation of affective language that this
dissertation investigates.
1.2 Theories of emotion
Unfortunately, one of the most significant things ever said about emotion
may be that everyone knows what it is until they are asked to define it.
(LeDoux, 1996)
The quote above from prominent emotion researcher Joseph LeDoux highlights the
difficulties involved with defining emotion. Nonetheless, this thesis will develop a
working definition to make it clear how we conceive of emotion and how we
operationalise it in the studies reported on here. Taking his cue from emotion
scholars such as Damasio, Panksepp, Frijda, and Scherer, Van Berkum (Van
Berkum, in press) develops a definition of emotion that highlights core aspects of
emotion that should be of particular interest to psycholinguists. These aspects can
9
be grouped, broadly speaking, around two issues: 1) how emotions arise and 2)
what emotions are for.
One of the fundamental points about how emotions arise is that they are
referential. That is to say, emotions are caused by, and about, some internal or
external stimulus. What triggers emotions is the appraisal of a stimulus as relevant
to our concerns. Importantly, Van Berkum points out that the appraisal and the
resulting emotion can, and often does, play out without conscious awareness. A
crucial point for language comprehension research is that, while emotions have
ancient, evolutionarily shaped triggers (i.e., pain, loud noises, emotional facial
expressions of conspecifics etc.), they can be triggered by any stimulus that can be
coupled to something else through associative learning— this includes linguistic
expressions as triggers for emotion. In fact, almost everything we encounter in the
world around us is appraised as somehow emotionally relevant, whether this
evokes full‐blown, concrete emotions or weaker and less defined, but no less
important, affective evaluations.
In addition to being about something, emotions are also for something.
That is to say, beyond merely appraising something as relevant to our interests,
emotions are for doing something about the stimulus that evoked it and to ensure
our wellbeing. Not only are emotions for doing something, in many cases they are
about doing something now. Strong emotions can take control if your immediate
homeostatic wellbeing is threatened, for instance in the case of fear in the face of a
predator. Arguably, weaker affective evaluations can also steer your behaviour in
the immediate future in subtle ways. For instance, in a conversation with someone
you dislike, even at a subconscious level, you will probably find a way to end that
conversation sooner rather than later.
Emotions prepare us to act and make us act in the present. These
preparations come in a package of short‐lived synchronised changes in various
systems and include action tendencies, physiological changes, cognitive changes, as
well as actual behaviours. For example, if somebody says something that makes
10
you really angry, a set of synchronised changes will unfold. These responses will
possibly include, among other things, increased heartrate and sweating, the release
of stress hormones. Importantly, in light of the research presented in this
dissertation, it will also include exhibiting that emotion through facial expression
and likely also through body posture, clenched fists, involuntary shouts or cries etc.
Two final points must be made about emotions. First, emotions are not
necessarily conscious. While your anger in response to what someone has said may
or may not emerge in consciousness as subjective feeling, the action package of
synchronised changes will become active even if the emotion does reach conscious
feeling. The second point is that the appraisal of a stimulus as relevant and the
subsequent emotional response to it will leave traces in memory. As a result,
whatever it was someone said to make you angry will be more likely to anger you
again in the future. Not only that, your response to related stimuli, for instance
other things that same person says later, will also be influenced by the traces of
your anger stored in memory. Below is a working‐definition highlighting those
aspects of emotion that are crucial in light of our research.
Emotions:
Are triggered by the (often fully automatic) appraisal of stimuli as relevant
(biologically, culturally, and/or idiosyncratically) to one’s concerns,
including linguistic stimuli.
Evoke synchronised sets of short‐lived changes, including physiological
changes and behaviour (such as facial expression), aimed at making you
do something about the emotional stimulus.
Leave traces in memory that influence the response to the emotion‐
evoking stimulus as well as other stimuli related to it.
11
1.2.1 Classifying emotion
Beyond abstract definitions of what emotion is there are various ways in which
theorists and researchers talk about which emotions there are. Different attempts
at classification have been made, broadly speaking aiming to either categorise
distinct emotions or organise emotions along certain common dimensions. Some of
the earliest proponents of either view were Charles Darwin (1872/1998) and
Wilhelm Wundt (1977/1907), respectively.
Emotion classification following Darwin has resulted in what is commonly
referred to as the Basic Emotions Model. Basic emotions are commonly believed to
biologically hard‐wired and to be the psychological basis of other, non‐basic,
emotional states. One classic inventory of basic emotions is based on the
supposedly universal recognition and expression of emotion through facial
expression and comes from Ekman, Friesen & Ellsworth (1982): anger, disgust, fear,
joy, sadness, surprise. However, Ekman (1999) later extended this list of basic
emotions to 15 emotions. Turner and Ortony (Turner & Ortony, 1992) made an
inventory and revealed how widely both the number and the nature of the
emotions included in lists of basic emotions varied. In addition, the criteria used
for inclusion differed greatly as well. Some of the criteria most hotly debated are
summarised by Shiota & Kalat (2012): basic emotions should be 1) universal, 2)
expressed in a distinct, built‐in way 3) arise early in life, and 4) be phsyiologically
distinct.
The dimensional approach to emotion classification takes a different tack.
Instead of assuming the existence of basic and discrete emotions, the dimensional
approach investigated the possibility that the subjective experience of our
emotions (i.e., what we call ‘anger’ or ‘joy’) can be arranged along more abstract
dimensions. The circumplex model (Shiota & Kalat, 2012) suggested two such
dimensions to account for the emotional space: valence (positive vs. negative) and
arousal (low vs. high). While the debate continues regarding whether valence
12
constitutes a single dimension, or positive and negative represent separate
dimensions (for a discussion, see Cacioppo, Berntson, Norris, & Higgins, 2012),
there is broad agreement that valence is a salient dimension that can meaningfully
organise emotions. Valence seems central to any emotion system (i.e., ‘is this good
or bad given the interests I am monitoring’) and offers a suitable starting point
given the aim of the research in this thesis. We will therefore bypass the
complexity involved with the status of certain emotions as basic, and simply
conceive of emotions in terms of valence. In addition to reducing theoretical
complexity, choosing valence as opposed to discrete emotions has methodological
benefits with regards to reliably measuring affective responses using facial
electromyography. These methodological considerations will be discussed next.
1.2.2 Measuring emotion using facial electromyography
Electromyography refers to the use of surface or intra‐muscular electrodes to
measure muscle action potentials. These myoelectric signals can be measured
reliably as an indicator of the activity of a specific muscle (Tassinary, Cacioppo, &
Vanman, 2000). We focus here on the use of surface electromyography used
specifically to measure activity of facial muscles in relation to affective stimuli.
There are a number of muscles in the face that are useful for inferring emotional
states (e.g., van Boxtel, 2010). It is important to note the advantage of facial EMG
in its ability to pick up muscle action potentials that do not result in visible facial
expression (Tassinary & Cacioppo, 1992).
As discussed in section 1.2.1, this thesis focuses on a valence distinction of
affective meaning rather than on discrete emotional constructs like surprise, fear,
sadness, anger, etc. Two muscles in the face are commonly used to distinguish
valence: the corrugator supercilii (frowning muscle) and the zygomaticus major
(smiling muscle). How activity of the corrugator and zygomaticus muscles
corresponds to stimulus valence relates to a basic physiological difference between
13
the two muscles. There is significant difference between muscles with regards to
their innervation ratio (Tassinary, Cacioppo, & Vanman, 2000). The innervation
ratio refers to the number of muscle fibres that are innervated by a single motor
neuron. The higher the innervation ratio (i.e., a single motor neuron activates a
larger number of muscle fibres) the stronger and more enduring the resulting
muscle activity will be. Muscles with a higher innervation ratio are also known to
tire less quickly. The corrugator is a muscle with a relatively high innervation ratio
and is characterised by more‐or‐less constant baseline activity. Compared to this
baseline, corrugator EMG activity reliably increases in response to negative stimuli
and decreases in response to positive stimuli. The zygomaticus, on the other hand,
has a much lower innervation ratio and is characterised by short‐lived, more
precise activity. Unlike the corrugator is has no enduring baseline activity and is
characterised by increased activity in response to positive stimuli and the absence
of a response to negative stimuli.
In a systematic study of affective pictures, sounds and words Larsen,
Norris & Cacioppo (2003) confirmed that corrugator activity reliably increased in
response to negative stimuli and decreased in response to positive stimuli. The
zygomaticus, on the other hand, did not respond to negative stimuli and increased
in activity in response to positive stimuli, making its mapping onto valence less
informative. In addition, the unambiguous interpretation of zygomaticus ‘smiling’
activity as reflecting positive emotion is problematic when we consider that we also
smirk sardonically, grimace, and offer wry smiles. In sum, the corrugator offers a
reliable, informative, and unified measure of valence and will therefore be the
focus of the research presented in this book.
14
3.1. Language comprehension and narrative
As the example of Lolita illustrated, the potential conflict between the affect
referred to by language and our affective evaluation thereof is easily enacted with
a narrative. The studies reported in chapters 2‐4 use the narrative format to
operationalise this affective complexity and investigate the interaction between
language‐driven simulation and evaluation as indexed by online corrugator activity.
Considering the importance of narratives to this research enterprise, it must be
made clear what narrative is. The following section will draw on established
definitions to formulate our working definition of narrative. In doing so, it will
become clear what narratives are and also why narrative is a suitable medium to
investigate our questions about how we process language describing emotion.
1.3.1 Defining narrative
There are many definitions of narrative and they differ in various ways, including
their point of departure. Some are formulated to be used in literary scholarship,
others for the study of media and communication, and others still with an
interdisciplinary aim to link psycholinguistics and narrative scholarship. A
comprehensive enumeration and comparison of definitions is beyond the scope of
this thesis, but despite their differences there is considerable overlap in definitions.
One core notion that most definitions agree on is that narrative involves the
representation of at least one event (e.g., Porter Abbott, 2008).
While seemingly trivial, the requirement for the presence of an event is a
good start to separate narrative from, for instance descriptive, expository, and
argumentative text. Beyond the insistence on the inclusion of one or more events,
the elements required by different definitions vary, but there is broad consensus
that events have to be ordered both in time (e.g., Herman, 2009) and logically, for
instance as cause and effect (e.g., Bal, 2009). Another point on which most
15
definitions agree is that a narrative must contain one or more agents (e.g., Ryan,
2007).
To illustrate, imagine a narrative about a game of chess. A list of chess
moves technically constitutes a number of events, but it does not tell us a story
about a game of chess. Ordering the moves as a sequence in time from first to last
helps, but is not sufficient. Despite having chronological order, each move could be
plucked randomly from a different game. By linking each move logically to every
other move we begin to see the outline of a narrative about a game of chess.
Without agents however, this list of causally related chess moves distributed
over time might as well be a description of a chain reaction of chemical reagents.
To make it a narrative we need agents that are human‐like in that they must be
possessed of intentionality. I say human‐like because such agents could just as
easily be anthropomorphised teacups. The discussion so far yields a provisional set
of necessary elements for narrative:
Representation of one or more events
A chronological structure of events
Logical relationships between events (i.e., cause and effect)
Human or human‐like agents
These basic elements arising from narratological approaches correspond well
to the event‐indexing model in a situation model account of (narrative) language
comprehension (e.g., Zwaan & Radvansky, 1998; Zwaan, Langston, & Graesser,
1995). A situation model refers to a mental representation, by the reader, of a
given discourse; it is a dynamic model that continuously integrates incoming
information into an ever richer micro‐world. The event‐indexing model is distilled
from a large body of empirical evidence that lists five salient dimensions along
which events in the discourse are continuously updated as part of the construction
a situation model:
16
Time
Space
Protagonist
Causality
Intentionality
Much of the initial research in the line of event indexing and situation models
relied on behavioural data. More recently, neuroimaging studies have shown that
neural activity tracks changes along the salient dimension from the event‐indexing
model while participants read a narrative (e.g., Speer, Zacks, & Reynolds, 2007;
Whitney, et al., 2009). This event‐indexing model has also been extended to the
processing of narrative film (Cutting & Iricinschi, 2015; Zacks, Speer, Swallow, &
Maley, 2010). The analytically derived set of basic elements of narrative thus
converges with a list of salient dimensions that have been empirically proven to be
a core part of not only narrative comprehension, but also more general event
comprehension.
1.3.2 Narrative as mimicry of experience
In addition to the elements outlined above, two additional elements help shed light
not only on what narrative is, but also what makes narrative so interesting to us as
humans. The first is particularity. Herman (Herman, 2009, pp. 89‐99) argues that
narratives tend to be particular rather than generic descriptions of events.
According to Herman, the degree of particularity describes a sliding scale between
explanation and narrative. That is to say, the more prototypical narrative text is
one that describes, in greater detail, a unique game of chess between specific
individuals in a particular time and place rather than an explanation of how a
generic game of chess might play out.
17
The second element is the inclusion of descriptions of the experience of
the agents involved; their thoughts, feelings, and perceptions. Herman refers to
this as qualia or ‘what it is like’ (Herman, 2009, pp. 21‐22) and Fludernik as
‘experientiality’ (1996, p. 12). What is meant by these terms is that the more
prototypical a narrative, the more likely it will include details about how our chess
players feel; sweaty hands, intense focus, excitement at a succesfull stratagem, and
happiness for the eventual winner.
Fludernik’s notion of experientiality as a central component to narrative
also involves the proposition that narrative mimics real human experience in that it
mirrors the basic ways in which we cognitively engage with the world around us
(1996, p. 35). The particularity of the events in combination with detailed
experiential information contributes to the vividness of this mimicry of human
experience. The vivid mimicry of a real experience is almost certainly one of the
things that draw readers to stories in the first place.
The idea that narrative mimics the way we understand the world around
us is taken further by Porter Abbott (2008, pp. 7‐12) by positing that we
fundamentally perceive the world as narratives. He illustrates this with examples of
how we automatically perceive narratives unfolding in static pictures or paintings.
Another illustration of our proclivity to perceive narratives where there are, strictly
speaking, none is a seminal psychology experiment (Heider & Simmel, 1944). Their
participants were tasked to describe a short animation that showed the
movements of geometric figures on a white field. Almost all participants described
seeing animate agents, usually humans, acting out a connected story. Only one
participant reported seeing geometric figures moving on a white background. The
notion of mimicry arising from narratology bears some similarity to the theory of
grounded cognition. Fludernik (2003) even explicitly bases her definition of
narrative on a cogntiviely grounded relationship between human experience and
the representation of that experience in narrative text.
18
The ingredients of a protypical narrative—including information about
what it feels like— the ubiquity of narratives and the way it reflects the way
humans perceive the world around them make narrative a rich and ecologically
valid arena to study affective language processing. In fact, I would argue that any
grounded theory of language comprehension that includes simulation would not be
complete without accounting for narrative comprehension. Because
comprehension is not a ‘cold’ cognitive process, and we actually care, sometimes
quite deeply, about what stories relate, we arrive back at the central question
regarding the role of simulation and evaluation in real‐time language
comprehension: is the corrugator activity (or another index of affective valence) we
see during online affective language comprehension the result of simulation or
evaluation? And moreover, what if our evaluation is at odds with what we ought to
be simulating?
This introduction began by illustrating this tension between simulation and
evaluation with a work of great literary esteem, but it can equally easily be
demonstrated using stories we tell each other every day. A prime example is the
ubiquitous practice of gossip as a tool of social control and reputation management
(Dunbar, 2004). For example, I might tell you a story descibing how happy a mutual
acquaintance was when they received their degree. Now, simulation and
evaluation of this happiness would align (i.e., they would both be positive) if you
know this mutual acquaintance to be an honest, hard‐working individual. However,
if you know them to have plagiarised a paper, defrauded exams, or you know they
cheated on your friend, simulation of this person being ‘happy’ would still have to
be positive, but your evaluation of their happiness would very likely be negative.
The next section develops a generalised framework for how this potential conflict
between simulation and evaluation could be reflected by corrugator activity.
19
Figure 1. Driver model for facial EMG activity during online affective language comprehension
1.4 A language processing framework for facial EMG activity
This introduction has thus far outlined the three critical components in the
investigation of affective complexity in real‐time language processing: 1) the
theoretical framework of grounded cognition and the notion of simulation 2) how
we measure and operationalise emotion/affect, and 3) the nature and suitability of
narrative as a medium. Based on these three elements, this final section presents a
framework for facial EMG activity during real‐time affective language
comprehension. This framework includes a schematic of how the reader would
process a narrative text and identifies where and how simulation and evaluation
might be expected to activate motor control areas governig corrugator activity,
which is then recorded using surface facial EMG.
20
The dotted lines labelled S1 and S2 both refer to language‐driven simulation.
Driver S1 describes simulation of lexical concepts as they are retrieved. Drawing on
the example of the quote from Lolita once more, it would describe simulation of
lexical concepts such as ‘love’. Driver S1 is the most likely force behind the effects
previous affective language processing studies have found using individual
affectively salient words (e.g., Foroni & Semin, 2009; Künecke, Sommer, Schacht, &
Palazova, 2015).
Driver S2 describes simulation as part of a situation model, a mental model
of, for instance, Humbert Humbert experiencing the feeling of being in love as
described in the quote at the start of this chapter. Driver S2 will likely have
conributed to the effects found in experiment using longer fragments of language
in facial EMG studies (e.g., Glenberg, Webster, Mouilso, Havas, & Lindeman, 2009;
Havas, Glenberg, Gutowski, Lucarelli, & Davidson, 2010; Fino, Menegatti, Avenanti,
& Rubini, 2016).
Driver E for evaluation, indicates the influence of the reader’s evaluation
of the situation model on the facial muscles. This is the factor that, to our
knowledge, none of the studies cited above that use facial EMG to explore
simulation effects during affective language processing have considered. In the
next three sections three possible ways in which the interplay between simulation
and evaluation might be predicted to play out in facial muscle activity, specifically
of the corrugator, as measured using EMG. Finally, the question mark represents
the empirical research question at the centre of this thesis: which of these possible
input drivers for corrugator motor control systems will surface in the actual
corrugator activity as measured by us using surface EMG?
1.4.1 Simulation‐only model
The simulation‐only model for corrugator activity during online affective language
comprehension predicts that corrugator activity will be solely determined by
language‐driven simulation (drivers S1 and S2). This model further predicts that any
21
affective response based on the evaluation of the affectively‐salient language will
not show up in the corrugator activity (driver E). Taking the concrete example of
the quote from Lolita at the start of this introduction, this would mean that
corrugator activity will be solely determined by the simulation of concepts like
‘love’ or the situation model containing the character, Humbert Humbert, who is in
love. Our negative evaluation of the fact that this type of love is entirely
inappropriate considering that it applies to a child, will not be reflected in
corrugator muscle activity.
This model is highly unlikely, considering that an important social function
of narrative is gossip, and facial expressions there are crucial for social alignment
(e.g., Dunbar, 2004). Furthermore, moral evaluation is largely automatic (Greene,
2014) and, despite the fact that it concerns a fictional narrative in a lab
environment, it seems unlikely that language‐driven simulation would override
moral evaluation entirely. Existing evidence shows that moral considerations do in
fact influence langauge processing, even in a lab context (e.g., van Berkum,
Holleman, Nieuwland, Otten, & Murre, 2009; Leuthold, Kunkel, Mackenzie, & Filik,
2015). Although unlikely, it is postulated here because one could argue that this is
precisely the model that is assumed in the facial EMG studies cited in the previous
section that interpret facial muscle activity as indicative of simulation of affective
language.
1.4.2 Evaluation‐blocks‐simulation model
The second model under consideration predicts that evaluation will determine the
corrugator activity entirely (driver E), leaving no trace of any language‐driven
simulation (drivers S1 & S2). Concretely this would result in corrugator activity
indicative of negative affect following our negative evaluation of Humbert Humbert
falling in love with a young girl. There would be no trace of any positive simulation
of either the concept ‘love’ or a situation model of ‘somebody being in love’. The
absence of any evidence of simulation in corrugator activity would not necessarily
22
mean that simulation does not occur in emotion systems. As facial EMG activity is a
down‐stream effect of simulation in emotion systems in the brain, it is possible that
simulation does occur, but does not surface in fthe resulting corrugator activity
measured using EMG. This could be because evaluation overwhelms it or because
realisation of motor impulses resulting from simulation is blocked in these cases of
vicarious experience, just as muscle activity is inhibited in sleep during dreams
(e.g., Schenk, 2015).
1.4.3 Multiple‐drivers model
The third possible model is one where both simulation and evaluation leave traces
in corrugator activity during online affective language comprehension. The
contribution of both simulation and evaluation could either occur concurrently or
successively. If they occur successively, one after the other, then a very plausible
arrangement is one in which language‐driven simulation first drives corrugator
activity and then is later overridden by evaluation. To return to our example, we
might first see positive affect reflected as a result of a simulation of the positive
affect described in the quote about being in love. This would then be followed by a
negative affective response when our negative evaluation takes over.
If the two drivers have a concurrent influence they would exert opposing
forces on the corrugator. This could result in an attenuated response of the
corrugator. In order to ascertain any attenuation, the corrugator response would
have to be compared to an instance where simulation and evaluation were not at
odds with each other. This is precisely what has been done in the three studies
reported in the next three chapters. Each of these studies builds on the next and
together they contribute to the proposed model for affective responding during
online language comprehension. Chapter 5 will consider the evidence from the
three experimental studies in relation to the proposed processing model and
discuss the implications for the various theoretical domains upon which this
research touches.
23
2. Study 1
Emotion in Stories: A Facial EMG Study on Simulation vs. Moral Evaluation
2.1. Introduction
Imagine this: you walk up to your car and see a kid scratch it with a key
and run off. There is a good chance that you would be furious and frown at least a
little bit as part of the expression of that anger. Facial expression is part of our
emotional evaluation of the world around us (Darwin, Ekman, & Prodger,
1872/1998; Keltner & Ekman, 2000). Because we reliably frown more when we
evaluate things as being negative and less when we deem something positive, the
corrugator supercilii muscle is especially useful as an indicator of our emotional
evaluation of things, and of how we express that evaluation to others. Using
surface facial electromyography (EMG), we can accurately record corrugator
activity. Its strong negative linear relationship to emotional valence, ranging from
positive to negative, makes it a reliable indicator of the emotional significance of a
given stimulus (Larsen, Norris, & Cacioppo, 2003; Tassinary, Cacioppo, & Vanman,
2000).
Now consider the following sentence: “Mark is furious when he walks up
to his car and sees a kid scratch it with a key and run off.” A number of studies have
shown that simply processing affectively salient language, even if it describes some
24
fictional character’s anger at their fictional car being keyed, will also evoke
corrugator activity (e.g., Foroni & Semin, 2009; Glenberg, Webster, Mouilso, Havas,
& Lindeman, 2009; Niedenthal, Winkielman, Mondillon, & Vermeulen, 2009).
Within the framework of theories of grounded language processing, activity of this
facial muscle is believed to result from language‐driven simulation of what is being
referred to. Simulation, in these cases, is taken to denote the neural reactivation of
experiential traces stored from earlier perceptual, affective, and motor experience
with the world (e.g., Barsalou, 2008). Such simulation would be part of the retrieval
and semantic combination of word meanings (retrieving the semantics of furious,
and constructing the semantics of “Mark is furious”) and/or of building a situation
model (imagining Mark being furious; cf. Zwaan & Radvansky, 1998; Zwaan &
Kaschak, 2008). Within this framework, corrugator activity is most sensibly
interpreted as a downstream consequence of the neural simulation of affect
(rather than as reflecting a critical role for muscle activity itself in language
processing).
If the neural systems involved in real‐world emotional evaluation (e.g.,
your outrage over some situation or event) are also used as part of such language‐
driven simulation (e.g., reading about somebody else’s outrage over some situation
or event), this raises an interesting question: what happens when the same
systems are recruited to conflicting ends? The issue does not easily come up in
grounded language processing experiments that use only single words or short
sentences (e.g., “furious”, or “Mark is furious”), because, although such simple
materials afford simulation, they do not necessarily elicit a lot of evaluation.
Moreover, the evaluation that is elicited by such limited stimuli is likely to always
be congruent with language‐driven simulation. However, if we scale up complexity
by placing such materials in a richer discourse context (Van Berkum, 2008; Zwaan,
2014), evaluation may covary or conflict with simulation. Consider again the
example of reading about Mark’s car getting keyed; as we process the language
describing Mark’s anger and construct the associated situation model, neural
25
simulation in support of language comprehension will result in corrugator activity
that reflects negative affect. However, if we feel that Mark is a bad person, we may
well evaluate his negative affect as something he deserved, so as positive. While it
is well‐established that Schadenfreude does in fact occur in such cases (e.g.,
Feather & Nairn, 2005; Leach & Spears, 2009; Singer, et al., 2006) and that this can
also influence facial muscle activity recorded using EMG (Cikara & Fiske,
Stereotypes and Schadenfreude: Affective and physiological markers of pleasure at
outgroup misfortunes, 2012), to our knowledge no research has been done on how
such affective evaluation meshes with language‐driven simulation. Our experiment
focuses specifically on how the conflicting demands that affective evaluation and
language‐driven affective simulation make on emotion‐relevant motor systems
play out, and how that is reflected in the corrugator activity.
In our experiment we explored the potential conflict between language‐
driven simulation and emotional evaluation by measuring corrugator activity over
two critical segments in short narratives. First, we manipulated the moral status of
a main character by having this protagonist act either morally or immorally.
Second, we manipulated a critical event where the protagonist experiences
something that is positive or negative to them. The character morality
manipulation was designed to render the character moral (‘good’) or immoral
(‘bad’) to the average reader, and to as such create the basis for a differential,
character‐dependent moral evaluation of subsequent good or bad critical events
befalling the character. We assumed that good events happening to moral
characters would primarily be evaluated as fair and lead to corresponding positive
emotions (e.g., a sense of justice), whereas bad events happening to those
characters would primarily be evaluated as unfair and lead to corresponding
negative emotions (e.g., moral indignation, anger, pity). Critically, we assumed that
how readers evaluate those same events would change dramatically when these
events would befall immoral characters, with good events happening to those
characters primarily evaluated as unfair, leading to related negative emotions (e.g.
26
moral indignation, anger, irritation), and bad events happening to those characters
primarily evaluated as fair, leading to positive emotions (e.g., Schadenfreude, a
‘serves‐you‐right’ sense of justice). In Table 1 below, an example narrative
illustrates the design of our study, as well as the specific timing of presenting the
various fragments.
Baseline Neutral distractor image 3 s
Introduction
Mark is driving through the pouring rain, on his way to his
mother. He’s still in the inner city and big puddles have
formed. It’s been raining non‐stop since yesterday. Some
streets are practically flooded. There are few cars on the
road and fewer bicycles and pedestrians still. Mark is
headed for a giant puddle and spots a pedestrian on the
sidewalk.
18
s
Character
Morality
(moral/immoral)
Mark slows down to
avoid the puddle,
making sure he
doesn’t splash the
pedestrian.
OR
Mark accelerates
through the
puddle on
purpose to create
a big splash and
soak the
pedestrian.
5 s
Continuation
Once outside the city he is driving along on the freeway.
There still isn’t a lot of traffic and Mark is enjoying the
landscape and the drive. He’s got the radio on full blast and
sings along loudly. When he glances at the dashboard to
adjust the channel he spots a warning light. He forgot to
put petrol in the car and has been running on empty for a
while.
15
s
Critical Event
(positive/negative
for the character)
Mark is happy when
he immediately
spots a petrol station
and he avoids being
stranded.
OR
Mark is frustrated
when there isn’t a
petrol station in
sight and he
becomes stranded
by the roadside.
5 s
Press ‘space’ to continue to the next story
Table 2.1. Example narrative, also illustrating trial structure and timing
27
On the assumption that participants are in a neutral or otherwise
moderate affective state as they read the introduction, our predictions for
corrugator activity at the subsequent character morality segment are
straightforward: increased corrugator activity (frowning) for immoral actions, but a
decrease for moral actions (relaxation). Although also of interest in itself, such a
differential response would above all suggest that the character morality
manipulation was successful, and thatthe stage would be set for subsequent
events.
The critical predictions concern what happens when readers subsequently
read about good or bad events befalling the character. With moral characters,
negative events are expected to clearly increase corrugator activity, because of a
negatively valenced simulation of the character’s state, a negatively valenced
evaluation by the reader, or a combination of the two. For similar reasons, positive
events befalling moral characters are expected to relax the corrugator, because of
a positively valenced simulation, a positively valenced evaluation, or both. With
immoral characters, however, simulation and evaluation valence will conflict, at
least for the average reader: something bad happening to an immoral character is
negative for the character but positive for the reader, and something good
happening to an immoral character is positive for the character but negative for the
reader. Hence, with immoral characters, language‐driven simulation should in
principle recruit the neural systems that control the corrugator to simulate one
valence (positive or negative) while evaluation should in principle recruit those
same systems to express the opposite valence. The outcome of this conflict is the
main focus of the experiment.
We considered three possible models to predict corrugator activity at the
critical event. The first is one where, during narrative fiction reading, language‐
driven simulation totally captures the corrugator, and does not reflect evaluation.
This simulation‐only model predicts that even in the face of oppositely valenced
evaluation (i.e., good or bad things befalling an immoral protagonist), corrugator
28
activity would simply show the simulation involved in constructing a situation
model (imagining the protagonist as frustrated or happy) and/or the simulation of
lexical meaning (“frustrated”, “happy”) in the service of such construction. We
deem this model rather unlikely, in part because of the evolutionary significance of
narrative in moral and other social‐affective evaluations (notably in gossip, Dunbar,
2004), and because of the roles that affective evaluation and the accompanying
overt facial expressions play in socially responding to interpersonal narrative (with
facial expression usually seen as a constituent of such emotion, see Van Berkum, in
press). Furthermore, as simulation is itself grounded in such primary affective
evaluation, it would be peculiar to predict that in language comprehension,
simulation prevails over ‘the real thing’. We include the simulation‐only model as a
logical possibility, though, in part because it can be taken to represent the tacit
assumption made in much grounded language processing research.
The second model under consideration holds that, in line with the role of
narrative (e.g., gossip) in moral evaluation, the emotional response to the
perceived fairness of the event completely captures the corrugator such that the
neural systems controlling it are no longer available for language‐driven simulation.
This evaluation‐blocks‐simulation model predicts that the moral status of the
protagonist determines the ultimate evaluative valence of a critical event in terms
of fairness, causing the corrugator responses to critical events to ‘flip’ when those
events befall immoral rather than moral characters. For instance, relative to
positive events, negative events should lead to increased corrugator activity (i.e.,
negative affect) when they happen to a morally good character, but to decreased
corrugator activity (i.e., positive affect) when they happen to a bad character. This
model makes the reasonable assumption that, although the neural systems that
control facial muscles might be free for simulating emotion as long as people don’t
care about what they read, those systems will immediately be recruited in the
service of real evaluative emotion as soon as a narrative describes something
worthy of evaluation.
29
The third model we consider is one in which evaluation and simulation
both determine corrugator activity. This multiple‐drivers model predicts that
simulation and evaluation both leave traces in the activity of the corrugator as
indexed by EMG. Because the drivers may interact in ways that are difficult to lay
out in advance, the exact pattern of results is hard to predict. The one clear
prediction, though, is that the corrugator EMG patterns cannot be explained in
terms of one of the simpler models laid out before.
The strength of the effects hypothesised under these three models could
arguably vary as a result of individual differences. We included two questionnaires
that measure the tendency of people to experience emotions in response to
narratives and their tendency to empathize and/or sympathize with the emotions
of others. For the former we use the Transportability questionnaire (Dal Cin, Zanna,
& Fong, 2004), measuring the degree to which people are readily transported into
narratives. Research shows that transportation influences, among other things, the
intensity of our emotional experience of a narrative (e.g. Green, Brock, & Kaufman,
2004). The second questionnaire, the Adolescent Measure of Empathy and
Sympathy (AMES; Vossen, Piotrowski, & Valkenburg, 2015)2, measures three
components of pro‐social emotion: affective empathy (‘feeling what another
feels’), cognitive empathy (‘understanding what another feels’), and sympathy
(‘feeling for the other’).
2 We opted to use the AMES because it neatly separates cognitive and affective empathy, in a more principled way than alternative empathy questionnaires (see Vossen et al., 2015, for discussion). Although originally targeting adolescents, the items that make up the scales were not deemed inappropriate for our age group.
30
2.2. Method
2.2.1 Participants and Stimulus Material
60 students (47 female, age range 18‐30, M = 21.08, SD =3.43) recruited from the
participant pool of the UiL OTS participated in exchange for financial compensation
(€ 12,‐). Before the experiment started, participants read and signed an elaborate
informed consent form (available from the corresponding author upon request)
detailing the nature of the materials and the procedure as well as emphasising
participants’ right to withdraw consent at any time during the experiment without
being required to provide a reason or losing the right to compensation. All
participants were native speakers of Dutch, without a diagnosis of dyslexia, without
Botox injections in the face and with normal or corrected‐to‐normal vision.
64 Short narratives were created for the experiment according to the
structure outlined in Table 1 above, each in four variants based on our 2 x 2
(morality x event) design. The character morality manipulation was pre‐tested in a
different group of 38 students (35 female), recruited from the same participant
pool and similar to the experiment group. The pre‐test participants were divided
into two groups and each read half (32) of the stories up to and including the moral
manipulation. They were first asked to rate, on a 7‐point scale, how prosocial (1) or
antisocial (7) the actions of the protagonist were, and, second, how expected (1) or
unexpected (7) the actions of the protagonist were. Moral actions were considered
more prosocial (M = 1.69, SD = 1.04) than immoral actions (M = 5.99, SD = 1.04).
Moral actions were also considered slightly more expected (M = 3.17, SD = 1.57)
than immoral ones (M = 5.16, SD = 1.56).
We created four pseudorandomized stimulus lists with 64 narratives each,
such that (a) every narrative occurred once in one of four variants in each list, (b)
participants would see 16 narratives in each of the four conditions, 8 with a male
and 8 with a female protagonist, (c) average item properties in each list were
similar in terms of pro‐sociality (and expectedness as its inevitable correlate(d) two
31
lists were in the reversed order, and (e) each narrative occurred with both male
and female protagonists across the four different lists, with the exception of 9
narratives that had fixed gender due to stereotypical behavioural expectations.
Dutch stimulus materials are available upon request (see Supplementary
Information A1).
2.2.2 Procedure and Data Acquisition
After reading and signing an informed consent form, participants received verbal
instructions. In a separate room, stimuli were presented on a screen at a distance
of approximately 60 cm. Stimuli were presented as outlined in Table 1, in white
Times New Roman (24 pt.) on a black background. Participants read 64 narratives
in total, preceded by 2 practice trials. Presentation rate of trials was self‐paced with
two fixed longer pauses during the experiment. Facial EMG activity was measured
using reusable Ag/AgCl electrodes with a 2 mm contact area over corrugator and
zygomaticus muscles on the right side of the face (van Boxtel, 2010). Raw EMG
signals were recorded with a NeXus‐10 MKII biosignal system (Mind Media) at a
sampling rate of 2048 Hz. After finishing this part of the experiment, electrodes
were removed, participants moved to a laptop to fill out the questionnaires, and
finally received their payment.
2.2.3 Data Preparation and Analysis
EMG data. The raw data were band‐pass filtered between 20‐500 Hz (48 dB/octave
roll‐off) and were additionally filtered with a notch filter at 50 Hz (see Van Boxtel
2010 for justification of filter parameters), followed by signal rectification and
segmentation per narrative, all in BrainVision Analyzer 2. For each narrative, the
3000 ms before story onset, consisting of the presentation of a neutral distractor
32
image of a forest scene, were inspected for remaining artefacts. We selected
maximally long continuous baseline epochs, with the requirement of a minimum of
500 ms of artefact‐free signal for both muscles simultaneously. Trials were
excluded if such a 500 ms baseline epoch could not be found (data loss of 0.50%).
We included the zygomaticus (smiling) muscle to afford compatibility with
previous studies addressing emotional valence. However, as Larsen, Norris, &
Cacioppo (2003) note, the zygomaticus is primarily a marker of positive affect only
and thus does not reflect emotional valence in the same way the corrugator does.
The same study also showed that zygomaticus was less reliable as an indicator of
valence in the case of linguistic stimuli. We would add to this that, especially in
more complex environments such as our narrative stimuli, smiling activity may be
difficult to interpret: smiles can be wry, sarcastic, and smirking as well as
expressions of true positive feeling. We therefore focus on the corrugator, and
report the zygomaticus data in Supplementary Information D1 for reference.
Following baseline selection, data were exported to MatLab and then
further segmented in two epochs of 5000 ms, time‐locked to the onsets of the
character morality and critical eventsegments in each story. Next, the data for both
the character morality segment and the critical event segment were divided into 50
consecutive 100 ms bins for optimal temporal resolution and reduction of random
error, with the average EMG response in each bin expressed as a percentage of the
pre‐story baseline epoch (expressing responses as a percentage of baseline helps
reduce random variance both within and between individuals; van Boxtel, 2010).
Supplementary Information B1 shows continuous average activation in 100 ms bins
for an entire trial.
We performed a Mixed Models Linear Regression analysis (using SPSS
version 24) for both critical segments. Rather than simply looking at average
activation over the whole segment, we built a growth curve model that also
captured linear, quadratic and cubic trends in the signal (Peck & Devore, 2008;
Mirman, 2015; trend components were centred to avoid correlation between trend
33
components). Using three trend components gives us two flex points and allows us
to describe a response in some detail while retaining interpretability and avoiding
over‐fitting (Mirman, 2015). Models were fitted with 100‐ms resolution, but for
ease of comprehension, parameter estimates (e.g., a b for a linear slope) will be
reported per second. Note that while we do not report effect size metrics, our
parameter estimates reflect percentages and as such already give an indication of
the size of the effect of a given predictor. We included separate trend components
for each condition and added these iteratively, so as to achieve maximal modelling
flexibility without forcing the model to fit, for instance, a quadratic trend for all
conditions when only some contained a significant quadratic component. The
model included subjects and items over lists as random intercepts, and random
slopes for the trend components on the subject factor, but in the Results we only
report (comparisons on the) linear component. We added predictors iteratively and
used the −2LL chi‐square test of model fit to assess whether each added factor
improved the model (p < .05). Supplementary Information C1 contains the
complete model summaries.
Questionnaire Data. For the transportability questionnaire, we calculated a basic
total score ranging from 20 to 180. The AMES questionnaire contained three
subscales each ranging from 1 to 5: affective empathy, cognitive empathy, and
sympathy. We calculated the averages for each subscale based on the original
authors’ validation study (Vossen, Piotrowski, & Valkenburg, 2015). To analyse the
individual differences, we used the mixed models procedure again, but without the
growth curve to avoid overcomplicating the analysis with four‐way interactions. We
examined the effect of each trait on average corrugator activation over the entire
5000 ms of each of the manipulated segments, entering the individual differences
as covariates in the fixed part of the model.
34
2.3. Results
2.3.1 Character Morality Manipulation
As can be seen in Figure 1, reading about moral and immoral actions elicited a
clearly differential corrugator response.3 For moral actions we saw a gradual and
modest decrease in activity. In contrast, we found a rapid and substantial increase
in corrugator activity starting within the first second in response to immoral actions
of the main character. Statistical analysis corroborated these observations. First,
the fixed effects revealed an effect of character morality, with immoral actions
leading to higher activation than moral actions (difference b = 29.46, t (26.42) =
13.02, p < .001, 95% CI of [25.00 33.91]).34 Furthermore, whereas moral actions
induced a significant linear decrease in corrugator activity (b = −1.79, t (59.99) =
−4.02, p < .001, 95% CI [−2.69, −0.90]), the regression for immoral ac ons included
a significant linear increase in corrugator activity (b = 15.14, t (59.97) = 4.55, p <
.001, 95 % CI [8.48, 21.80]), with both linear trends also differing significantly from
each other (p < .001; all slope estimates per second). This is in line with our
predictions that moral actions would elicit positive affect and immoral actions
would elicit negative affect. It also suggests that the stage was effectively set for
our subsequent critical event manipulation. For a complete report of the results,
see Supplementary Information C1.
3 The corrugator responses in Fig. 1 start slightly above the 100% baseline level. We suspect this reflects the additional effort demanded by reading the preceding introduction segment of the narratives, in comparison with passive viewing of the preceding neutral image. 4 Because our growth curve analysis used centred time components, the intercept in our analyses represents the average activation over the entire time window, rather than the point where the curves leave the Y‐axis. Furthermore, whereas the reported intercepts are based on fixed effects in the final model only, the fitted line in Figure 1 includes both the fixed effects and the random effects from the model.
35
Figure 2.1. Observed averages of corrugator response during character morality
segment, with growth curve model regression overlaid.
2.3.2 Critical Event Manipulation
Figure 2 below shows the results for the critical event manipulation: something
positive or negative befalling the character. Our predictions concerned the effect of
the character morality on the valence of the corrugator response, in relation to the
valence of the event. We will therefore discuss the critical events separately for
moral and immoral characters.
36
Figure 2.2 Observed averages of corrugator response to critical events befalling
moral and immoral characters, with growth curve model regression overlaid.
Moral Characters. The corrugator response to critical events happening to moral
characters revealed a clear differential pattern for positive and negative events
(see Figure 2, squares and solid lines), with negative events eliciting an increase in
corrugator activity, and positive events leading to a decrease in activity. This
pattern was corroborated by statistical analysis. For moral characters, negative
events elicited significantly stronger mean corrugator activation than positive
events (moral‐negative ‐ moral‐positive difference: b = 21.11, t (272.33) = 8.19, p <
.001, 95% CI of [26.18, 16.03]), and also resulted in clearly different temporal
developments. In particular, whereas the model regression for moral‐negative
contained a linear increase in activation (b = 3.60, t (95.84) = 3.20, p = .01, 95% CI of
37
[1.37, 5.84]), moral‐positive conditions led to a linear decrease in activation (b =
−2.58, t (59.74) = −4.54, p < .001, 95% CI [−3.71, −1.44]). Note, all slope es mates
are reported per second. A pairwise comparison of these linear trends also revealed
they differed significantly at the p < .001 level.
Although these different corrugator results reflect a difference in valence
of the critical event, the data at this point do not allow us to say whether the
corrugator response here reflects language‐driven simulation, moral evaluation, or
a mixture of the two. This is because in the case of moral characters, simulation
and evaluation predict the same valence for the corrugator response. To illustrate,
language describing a negative event befalling a moral character would involve
simulation of negative concepts or negative character emotions in the situation
model and as such lead to increased corrugator activity. By the same token
however, the evaluation of a negative event befalling a good character as unfair
would also lead to increased corrugator activity. Our design included the immoral
character conditions to pull these potential drivers of the corrugator response
apart.
Immoral Characters. The corrugator response for immoral‐positive and immoral‐
negative conditions (see Figure 2, triangles and dashed lines) presented a very
different picture from that for the moral characters. At first glance, the immoral
conditions did not show a clear valenced response at all. The statistical model
confirmed this impression. With immoral characters, mean corrugator activity did
not significantly depend on whether a positive or negative event occurred
(immoral‐positive ‐ immoral‐negative difference: b = 1.37, t (272.40) = 0.53, p = .60,
95% CI [−3.71, 6.44]). Furthermore, although including linear and quadratic
components for both positive and negative events befalling characters significantly
improved the model, none of the linear and quadratic estimates themselves
significantly different from zero (see Supplementary Information C1). In all, the
38
statistics indicate a flat‐line corrugator response in both cases and no significant
difference between the two.
These results do not sit well with the simulation‐only model (which
predicted similar corrugator effects of negative versus positive events regardless of
protagonist status), nor with the evaluation‐blocks‐simulation model (which
predicted that the pattern of corrugator activation would track fairness, and hence
flip for bad protagonists, relative to good protagonists). Because the results cannot
easily be explained in either of these simple models, they suggest that some
version of the multiple‐drivers model, where both language‐driven simulation and
moral evaluation exerted control over the corrugator, is appropriate here. We will
explore this further in the Discussion section.
2.3.3 Individual Differences
We analysed the individual differences by entering each personality trait
as a continuous covariate in the fixed part of a simplified model which included
morality and critical event, but excluded linear, quadratic and cubic trends (see
Methods). The personality traits included Cognitive Empathy (M = 3.54, range =
2.17‐4.83) Affective Empathy (M = 3.01, range = 1.60‐4.00) Sympathy (M = 3.86,
range = 2.5‐4.75) and Transportability (M = 73.19, range = 39‐132). Our models
revealed that in either of the critical segments, only Affective Empathy and
Transportability significantly modulated corrugator activity (see Supplementary
Information E1).
At the character morality segment, Affective Empathy scores only had an
effect on corrugator activity elicited by immoral actions (b = 16.16, t (61.74) = 3.46,
p < .001, 95% CI [6.84, 25.48]), with higher Affective Empathy scores associated
with more negative affect. At the same segment, Transportability also modulated
the corrugator responses elicited by immoral actions only (b = −0.55, t (62.05) =
−3.42, p < .001, 95% CI [−0.87, −0.23]), with higher Transportability scores
associated with less negative affect. Affective Empathy and Transportation thus
39
seemed to have the opposite effect on frowning activity during immoral actions.
Because Affective Empathy and Transportation scores themselves were negatively
correlated (r(58) = −.17, p(two‐tailed) < .001), their impact on corrugator activity
may not be independent. Corrugator activity to moral actions did not show a
significant effect of either Affective Empathy scores (p = .40) or Transportation
scores (p = .54).
At the critical event segment, we found that in the moral‐negative
condition (bad things happening to good people) higher Affective Empathy scores
once again resulted in more corrugator activity (b = 9.17, t (62.72) = 2.36, p = .02,
95% CI [1.39, 16.95]), indicating that those higher in Affective Empathy clearly
exhibited more negative affect in response to the character experiencing a negative
emotion. Furthermore, we found the same relationship between corrugator
activity and Affective Empathy for the immoral‐positive condition (b = 8.61, t
(62.78) = 2.21, p = .03, 95% CI [0.82, 16.39]). As both conditions can be considered
to involve an evaluation in terms of unfairness, a parsimonious explanation might
be that those predisposed to feel what others feel display more negative affect
particularly in cases of unfairness, an account that can also explain the Affective
Empathy effect on corrugator responding to the first immoral action of the
character. At the critical event segment, Transportation scores did not co‐vary with
corrugator activity in any of the conditions (see Supplementary Information E1).
2.4. Discussion
Our study pitted language‐driven simulation against moral evaluation using
narratives containing a protagonist manipulated to be moral or immoral. We used
EMG to measure the response of the corrugator muscle as an indicator of these
two processes. For the character morality manipulation (e.g. “Mark slows down /
accelerates…”), we expected that the corrugator response would clearly reflect
moral valence. This prediction was borne out by our results: relative to a pre‐story
baseline, participants frowned more at immoral actions and less at moral actions,
40
with the difference between the two emerging relatively rapidly, in less than a
second after presentation of the critical sentence. This adds to existing evidence
that facial EMG recordings in general, and that of the corrugator in particular, can
help track responses to affectively loaded language (e.g. Foroni & Semin, 2013;
Foroni & Semin, 2009; Niedenthal, Winkielman, Mondillon, & Vermeulen, 2009;
Glenberg, Webster, Mouilso, Havas, & Lindeman, 2009), and extends that evidence
to a new domain, the processing of morally loaded language (e.g., Van Berkum et
al. 2009; Leuthold et al. 2015, for EEG indications of very rapid processing of such
language).
The character morality manipulation was crucial to untangle the two
possible drivers behind the corrugator response at the subsequent critical event
(e.g., “Mark was happy/frustrated when…”). Here, the presence of immoral
characters led to conflicting valence predictions depending on whether language‐
driven simulation or fairness‐based moral evaluation drove the corrugator
response, both in the immoral‐negative condition (a bad person befalls something
bad, i.e., negative for the character but fair in the eyes of the reader) and the
immoral‐positive condition (a bad person befalls something good, i.e., positive for
the character but unfair in the eyes of the reader). We considered three models of
how corrugator activity might reflect this conflict during the critical event
manipulation: (1) the corrugator response might simply track the valence of
language‐driven simulation of the event (simulation‐only model), (2) the corrugator
response would reflect only fairness‐based moral evaluation (evaluation‐blocks‐
simulation model), or (3) language‐driven simulation and moral evaluation would
both drive the corrugator response (multiple‐drivers model).
Because the observed pattern of corrugator activity radically depends on
the moral status of the protagonist, our data clearly refute a simple simulation‐only
model, according to which corrugator responses to phrases such as “Mark was
frustrated/happy” merely reflect generic situation‐model building (imagining the
protagonist as frustrated or happy) and/or simulating lexical‐conceptual meaning
41
(“frustrated”, “happy”) in the service of such construction. We already considered
this model rather unlikely, in part because socially important uses of narrative in
human exchanges, such as gossip, can only work if moral evaluation does not shut
down when people process language (Dunbar, 2004). Of course, when reading
fictional narratives in the lab, real social relations are not at stake and evaluative
responses therefore might well be attenuated. Given that moral evaluation is to a
large extent automatic (Greene, 2014), however, it is unlikely that it would be
eliminated in the lab. Our corrugator data at the critical event confirm this idea of
automatic moral evaluation, as does the large and rapid corrugator response to a
moral transgression earlier in the narrative.
Our results also refute the more plausible evaluation‐blocks‐simulation
model, which had predicted the corrugator responses to critical events to ‘flip’ as a
function of whether the protagonist had just behaved morally (stronger corrugator
activity for negative events than positive events) or immorally (stronger corrugator
activity for positive events than negative events). This is not what our data show:
whereas negative events lead to considerably more corrugator activity than
positive events when those events befell a morally laudable person, there was no
difference in corrugator activity between positive and negative events befalling
immoral characters. The results displayed in Figure 2 are thus best accounted for in
terms of a multiple‐drivers model, where both language‐driven simulation of events
and the evaluation of those events (i.e., the reader’s own emotions) simultaneously
recruit the neural systems driving the corrugator, at least with the materials
studied here.
How those two drivers interact exactly is as yet an open question. One
possibility is that in the case of immoral characters, the corrugator activations
associated with language‐driven simulation and moral evaluation cancel each other
out. In this version of the multiple‐drivers model, the corrugator increase due to,
say, simulating a frustrated protagonist and/or retrieving the associated lexical
semantics would then need to be levelled out by a corrugator decrease related to
42
the positive affect associated with evaluating this state of affairs as fair (e.g., a
‘serves‐you‐right’ sense of justice, Schadenfreude) in case this protagonist
previously behaved immorally. Likewise, with an immoral protagonist, the
corrugator decrease due to simulating his or her happiness and/or retrieving the
associated lexical semantics would then need to be levelled out by a corrugator
increase related to the negative affect associated with evaluating this particular
outcome as unfair (a sense of moral indignation, anger, etc.).5
Corrugator EMG has also been linked to ease of processing or mental
effort (e.g., Topolinski, Likowski, Weyers, & Strack, 2009; Cohen, Davidson, Senulis,
Saron, & Weisman, 1992; Cacioppo, Petty, & Morris, 1985) where enhanced
corrugator activity is indicative of increased effort and less fluency in processing. At
first glance, this may seem to be a potential confound for our results. However, the
immoral conditions are arguably the most complicated because evaluation in these
cases involves a reassessment of the valence of the event in light of the character’s
moral status. Yet, for these conditions we do not find any phasic increase in
corrugator actvity at the critical event. In fact, the only time corrugator EMG
increases during the critical event is when good characters experience a negative
event, arguably a fairly uncomplicated case. As such, complexity or disfluency in
processing does not offer a parsimonious explanation for the patterns we found
during the critical event segment.
An alternative account that we cannot as yet fully rule out relates to the
concept of identification. Identification with characters is often assumed to involve
the reader taking on the character’s goals and values as their own and experiencing
the emotions of the character (Oatley 1995). This could be reframed in terms of
selective, context‐dependent simulation at the level of the situation model, i.e. of
simulating the emotions of certain protagonists, but not those of others. There is
evidence for a connection between character likeability and identification (Chory
5 Although such exact cancellation may feel somewhat ‘coincidental’, we note that this outcome is as likely as any specific partial cancellation, such as when the impact of evaluation is twice that of the counteracting impact of simulation.
43
2013, Tian & Hoffner 2010) and reduced self‐reported experience of positive and
negative emotions in response to characters that readers identified with less
(Hoeken & Sinkeldam 2014). More generally, the emotional charge of the context
appears to affect both what and how much is simulated (Samur, Lai, Hagoort, &
Willems, 2015). If readers only simulate the emotions of likeable characters and not
those of disliked characters, a result such as in Figure 2 is conceivable. To explain
our corrugator findings for immoral characters with this ‘selective simulation’
account alone, however, one would also need to assume that readers do not
morally evaluate what they read or hear, at least not in the case of events befalling
immoral characters. With morally loaded narrative, and for reasons discussed
before, we deem such absence of moral evaluation highly unlikely. Also, more
generally, we find it unlikely that motor control programs that evolved to express
one’s own emotion are merely used to simulate other people’s emotions, and not
used to express ones own emotion over rather eventful social matters. If this were
the case, people would for example not be able to facially express, to each other,
their emotional alignment over a piece of gossip, or some other narrative about
what happened to them or others.
It would be useful to obtain more precise evidence on how the corrugator
response unfolds relative to crucial words in the unfolding sentence. Our critical
sentences were presented as a whole in our current study, so it is not exactly
known when critical words like frustrated or happy were read by the participants.
More precise time‐locking to words like happy or frustrated might thus reveal an
initial purely simulation‐driven corrugator response, rapidly followed by an
evaluative or mixed response. Furthermore, our critical events can be said to
consist of two components: an adjective which signals the valence of the emotional
state of the character (e.g., happy vs frustrated), and a clause describing the reason
for this state (e.g., when there isn’t a petrol station in sight and…). It is therefore
also conceivable that simulation‐driven responses dominate the first part, whereas
evaluation‐driven corrugator responses dominate the second part. Either way, a
44
more precisely time‐locked version of the same experiment with a more fine‐
grained control of when specific information becomes available might help to
differentiate the respective contributions of simulation and evaluation to the
unfolding corrugator EMG signal.
The effects of Affective Empathy suggest that the corrugator response to
morally loaded narrative is also subject to some individual variation. At the
character morality manipulation, readers with a higher Affective Empathy score
frowned more at immoral actions, and at the critical event manipulation, those
readers frowned more in response to bad things happening to good characters as
well as to good things happening to bad characters. A reasonable post hoc
interpretation is that people who are by habit or constitution more predisposed to
feel what others feel are also particularly responsive to unfairness, rather than
fairness. Furthermore, we found that readers with higher Transportability scores
responded less to descriptions of immoral actions than those with lower scores.
This could be taken to suggest that moral judgement of narratives decreases when
a reader is more immersed in the narrative, perhaps because of a shift in the
balance between evaluation and simulation. As Transportability correlated
negatively with Affective Empathy in our sample, these effects might not be
independent of each other. While these results provide food for thought, the
evidence is correlational, and must as such be approached with great caution.
In all, and independent of which precise variant of the multiple‐drivers
model will ultimately account for them, our results suggest that moral evaluation
has powerful effects on corrugator activity during narrative language processing:
sentences containing affective information such as “Mark is furious when…” and
“Mark is happy when…” generate quite different corrugator EMG responses as a
function of whether the protagonist has just displayed morally good or
objectionable behaviour. We take this result to reflect the reader’s direct and rapid
moral evaluation of, and associated emotional response to, what is being narrated,
both when reading about downright moral or immoral behaviour, and when
45
reading about events that befall the characters at hand. The implication is that
corrugator activity during language processing does not merely reflect simulation
of the protagonist’s emotion (e.g., Mark being frustrated) and/or of lexical‐
semantic meaning (e.g., retrieval of the meaning of frustrated). The fact that
people have emotions about other people’s emotions co‐determines how their
corrugator responds as they read a story about those other people.
Although the traces of affect picked up via facial EMG over the corrugator
may often not be visible in the face (Tassinary & Cacioppo, 1992), corrugator EMG
recordings show us that readers quite literally frown upon descriptions of
characters behaving immorally and that the moral status of characters drastically
influences corrugator activity during later affectively salient passages. This result
highlights the importance of unpacking coarse notions of affective meaning in
language processing research into components that reflect not only simulation but
also evaluation. Our corrugator EMG results here also call for a re‐evaluation of the
interpretation of corrugator EMG (and other affect‐related facial muscles) and
other peripheral physiological measures as unequivocal indicators of simulation in
affective language processing. Further exploration is needed of how such measures
behave in a richer and more ecologically valid language processing arena, such as
narrative. Such work would benefit the field by refining our understanding of the
role of simulation processes within a framework of grounded cognition in general
and language comprehension in particular.
46
47
3. Study 2
Temporal Development of Simulation and Evaluation of Emotions in Stories: Online processing of character affect
3.1 Introduction
One of the most enjoyable things about reading is that it allows us to walk
a mile in the shoes of characters from the most amazing stories. For example,
millions of readers have vicariously lived the life of the Machiavellian Queen Cersei
from the Game of Thrones book series6. This vicarious experience of ‘walking a mile
in another’s shoes’ is more than just a figure of speech. To illustrate, although the
reader will likely be sitting down as they read about Queen Cersei walking through
the palace gardens of the Red Keep, parts of their cortical (pre)motor areas usually
involved in walking will nonetheless be slightly activated (e.g., Aziz‐Zadeh, Wilson,
Rizzolatti, & Iacobini, 2006). Theories of grounded cognition hold that this is
because in order to understand what we read, we simulate the meaning of words.
Simulation involves the re‐enactment of multimodal states acquired during
6 The examples given in this paper do not contain spoilers about events from either the books or the TV show.
48
previous experience with the referents described in the language (e.g., Barsalou,
2008), either for the retrieval of the meaning of individual words or as part of
building a situation model of a longer text. Whether such simulation is automatic
and strictly necessary for comprehension has been debated (Mahon & Caramazza,
2008; Willems & Cassanto, 2011), but converging evidence supports the notion that
sensorimotor simulation is, at least under some circumstances, involved in
language comprehension (Kiefer & Pulvermüller, 2012).
While the example above deals with language referring to motor action,
simulation is also said to be involved in affective language processing, i.e., language
about, or otherwise relevant to emotion (e.g., Vigliocco, Meteyard, Andrews, &
Kousta, 2009). In particular, a large number of facial electromyography (EMG)
studies show that affective language processing is accompanied by congruent
muscle activation (e.g. Foroni & Semin, 2009; Glenberg, Webster, Mouilso, Havas,
& Lindeman, 2009; Havas & Matheson, 2013; Künecke, Sommer, Schacht, &
Palazova, 2015). These studies primarily rely on measuring activity of the
corrugator supercilii (frowning) muscle, which reliably reflects affective valence of
stimuli in a wide variety of input domains, including language stimuli (Foroni &
Semin, 2009; Glenberg, Webster, Mouilso, Havas, & Lindeman, 2009; Havas &
Matheson, 2013; Künecke, Sommer, Schacht, & Palazova, 2015). For instance,
when we read that Cersei is ‘frowning’ or that she is ‘angry’, our corrugator will
increase in activation. Conversely, if she ‘smiles’ or is ‘happy’, corrugator activity
will decrease. These changes in muscle activation do not necessarily result in
observable facial expression, but even in those cases we can still pick up the action
potentials in the muscles using surface facial EMG (Tassinary & Cacioppo, 1992).
Within the framework of grounded language comprehension this activity is
commonly interpreted to be a consequence of simulation in motor and emotional
systems in the brain.
However, in everyday language use, affective meaning is routinely more
complicated than the facial EMG studies cited above, and many others, allow (for a
49
discussion, see Van Berkum, to appear). This complexity becomes apparent if we
take the carefully controlled single words and short sentences typically used in
facial EMG studies on language processing and embed these inside a story. Imagine
reading ‘Cersei is furious when her favourite dress rips’ just a few lines after you
have read all about how Cersei ordered an innocent woman to be sent to prison,
feeling no remorse and even a certain measure of glee. While the word ‘furious’
still refers to a negative emotion, we may deem it less negative in light of what we
know about Cersei’s personality. Because as humans we are inclined to
Schadenfreude in the case of disliked or envied characters (Singer, et al., 2006;
Leach & Spears, 2009; Cikara & Fiske, 2012), we will probably evaluate Cersei’s
negative emotion as positive. This is because we do not merely understand
language dispassionately. Rather, just as we evaluate what we see, smell, or touch,
we evaluate what we read or hear and we care, sometimes quite deeply, about the
events described. And so we may ask, what will really be reflected in corrugator
EMG when, as in the example of the evil queen ripping her dress, the valence of
language‐driven simulation conflicts with the valence of our own evaluation of
what is described?
In a recent facial EMG study ('t Hart B. , Struiksma, van Boxtel, & Van
Berkum, under review), we explored this type of conflict by orthogonally
manipulating language‐driven simulation and moral evaluation within short
narratives. As in the example of Cersei above, we manipulated the moral status of
characters in narratives by first describing them as behaving either morally or
immorally in a story context. In a later segment of the narrative, these same
characters experienced a positive or a negative emotion in response to some event.
During the story segment where we manipulated character morality, corrugator
measurements revealed that participants quite literally frowned upon immoral
actions and relaxed when reading about moral actions. These results reflected
negative and positive affect in response to immoral and moral actions, respectively.
50
As for corrugator EMG while reading about subsequent ‘affective events’
befalling the same characters, we formulated three possible models of how
language‐driven simulation and evaluation might contribute to corrugator EMG
activity during the subsequent affective event befalling the same character.
According to the simulation‐only model, language‐driven simulation would be the
sole driver of the corrugator EMG response during the affective event. This model,
which reflects the current interpretation of facial EMG results in embodied
language processing research, simply predicts increased corrugator activity
(negative affect) when language describes a negative event for the character, and
decreased corrugator activity (positive affect) when language describes a positive
event for the character, regardless of the characer’s moral status.
The evaluation‐blocks‐simulation model holds that, in stories that are
sufficiently interesting to allow for evaluation, emotion‐relevant neural systems
controlling the corrugator are no longer available for language‐driven simulation,
and are fully recruited to reflect those evaluations. We assume that the salient and
dominant evaluation for our stimuli will be in terms of fairness. The evaluation‐
blocks‐simulation model predicts increased corrugator activity in response to
‘unfair’ events (negative events befalling moral characters and positive events
befalling immoral characters), and decreased activity to ‘fair’ events (positive
events befalling moral characters and negative events befalling immoral
characters).
The multiple‐drivers model, finally, allows for a combined influence of
language‐driven simulation and evaluation on corrugator EMG. For moral
characters, this combined influence would generate increased corrugator activity
to negative events (negative, and also unfair) and decreased activity to positive
events (positive, and also fair). For immoral characters however, simulation and
evaluation would counteract each other: negative events should involve simulation
of negative character affect, but such events are also evaluated as fair and as such
elicit positive affect (e.g., Schadenfreude), while positive events should involve
51
simulation of positive character affect, while being evaluated as unfair and as such
elicit negative affect (moral indignation). For situations describing the affective
state of immoral characters, these counteracting forces should lead to ‘dampened’
net corrugator EMG responses, i.e. smaller differences in the responses to negative
and positive events than in the case of moral characters.
Our results ('t Hart B. , Struiksma, van Boxtel, & Van Berkum, under
review) are compatible with the multiple‐drivers model. For moral characters we
found congruent responses to affective events: increased corrugator activity in
response to negative events and decreased activity in response to positive events.
For immoral characters, we found neither an increase nor a decrease in corrugator
activity as participants read about positive and negative events. In fact there was
no differential corrugator response to negative and positive events befalling
immoral characters at all. Because neither the simulation‐only model nor the
evaluation‐blocks‐simulation model accounted for the absence of a differential
response to valenced events befalling immoral characters, we concluded that the
multiple‐drivers model best accounted for this pattern; both simulation and
evaluation exerted control on the corrugator in such a way that, with our materials,
the two opposing forces cancelled each other out.
The results of our first study spoke against a processing model in which
only one of the drivers was behind the corrugator response during story reading.
However, although a multiple‐drivers model seems plausible, replication and
extension is in order. As for the latter, the previous study was limited in how we
presented the affective event stimulus. In all narratives, the affective event
consisted of a single sentence, presented all at once for 5 seconds: e.g., Mark is
frustrated when after a few minutes he runs out of petrol and becomes stranded by
the roadside. Although all sentences were of a similar length and structure, we had
no fine‐grained control over when participants were reading specific affect‐related
words (e.g., frustrated). This may have masked potential phasic effects of one or
52
both of the proposed drivers in response to specific affective information inside the
sentence.
In the current study we used the same narratives as ‘t Hart, Struiksma, van
Boxtel, & Van Berkum (under review), but presented the critical affective event
sentence in a piecemeal fashion, to optimise the chances of finding possible phasic
effects of language‐driven simulation and evaluation. As can be seen in Table 1, the
affective event contains two critical segments that are presented separately: an
‘affective state adjective’ describing the emotional state of the character, and an
‘affect reason segment’ detailing the event that led to the character’s emotional
state. Because an affective state adjective such as frustrated simultaneously
presents a highly focused trigger for simulation as well as – by revealing the
valence of the event for the character – for evaluation, without being confounded
by additional details on the event at hand, it represents the cleanest point at which
to assess the interaction of character affect and character morality in corrugator
EMG. Furthermore, by also time‐locking the EMG signal to the subsequent reason
for the character’s emotion, we have a second opportunity to examine the
interplay between simulation and evaluation (albeit in a way that is, due to the
varying nature of those reasons and their multi‐word description, somewhat less
precisely controlled).
53
Baseline Neutral distractor image 2 s
Introduction
Mark is driving through the pouring rain, on his way to his mother. He’s still in the inner city and big puddles have formed. It’s been raining non‐stop since yesterday. Some streets are practically flooded. There are few cars on the road and fewer bicycles and pedestrians still. Mark is headed for a giant puddle and spots a pedestrian on the sidewalk.
18 s
Character morality (moral/immoral)
Mark slows down to avoid the puddle, making sure he doesn’t splash the pedestrian.
OR
Mark accelerates through the puddle on purpose to create a big splash and soak the pedestrian.
5 s
Continuation
Once outside the city he is driving along on the freeway. There still isn’t a lot of traffic and Mark is enjoying the landscape and the drive. He’s got the radio on full blast and sings along loudly. When he glances at the dashboard to adjust the channel he spots a warning light. He forgot to put petrol in the car and has been running on empty for a while.
15 s
Transition … 1 s
Name
Mark
0.75 s
Verb
is
0.75 s
Affective state adjective
happy OR frustrated 1 s
Neutral
when after a few minutes
2.5 s
Affect reason
he spots a petrol station in time and avoids being stranded.
OR
he runs out of petrol and becomes stranded by the roadside.
2.5 s
Press ‘space’ to continue to the next story
Table 3.1. Example narrative illustrating trial structure and time on screen for each of 10 different segments. All segments were separated by a blank screen of 250 ms.
54
For the character morality segment we predicted substantially increased
corrugator activity (denoting negative affect) for immoral actions and a more
modest decrease in activity (denoting positive affect) for moral actions. As in our
previous study, this outcome would indicate that the manipulation of the
character’s moral status was successful. For the affective state adjective we
predicted, based on the multiple‐drivers model, that for moral characters the
combined influence of simulation and evaluation should generate increased
corrugator activity for negative state adjectives (negative and unfair) and
decreased activity for positive adjectives (positive and fair). For immoral characters,
we predicted that the difference between positive and negative state adjectives
should be smaller than for moral characters, because, under the multiple‐drivers
model, simulation and evaluation counteract each other.
Note that a multiple‐drivers model leaves open the possibility that
simulation and evaluation do not affect the corrugator at the exact same moment
in time. If simulation is crucial for comprehension, then purely language‐driven
simulation effects on the corrugator might actually emerge before the first
evaluation effects show up. We did not see any evidence for a brief ‘simulation‐
only phase’ in our earlier study, but any such effects may have been masked by the
relatively coarse sentence‐level time‐locking in that study. Because the current
study allows us to time‐lock the corrugator EMG response to the critical affective
state adjective, we are in a much better position to examine this possibility.
The affect reason segment provides an explanation for the character’s
emotional state described in the affective state adjective segment. The affective
valence here derives mostly from the meaning of the sub‐clause as a whole rather
than from particular words, making the time‐locking of the corrugator signal to
specific affective information less precise. We predicted that the corrugator EMG
patterns here would resemble those found in the previous study, but the
separation of the affective state adjective in the new design could also lead to
different patterns.
55
3.2 Method
3.2.1 Participants and Stimulus Materials
60 students (12 male) aged between 18 and 27 (M = 21.02, SD = 2.62) from the
Utrecht University Humanities Faculty participated in exchange for financial
compensation (€ 12). All were native Dutch speakers, without dyslexia and without
Botox injections to the face and with normal or corrected‐to‐normal vision.
We presented 64 narratives as outlined in Table 1. Each narrative had four
variants based on our 2 x 2 design (morality x affective event). The efficacy of the
character manipulations was pre‐tested and found to be successful in relation to
the previous study ('t Hart B. , Struiksma, van Boxtel, & Van Berkum, under review).
We generated four lists of 64 narratives for the main experiment. Each narrative
only occurred once in every list, with the four conditions distributed across lists.
Each list thus contained 16 narratives in each condition, 8 with a male and 8 with a
female character. All but 9 of the narratives occurred both with male and female
characters across different lists. We reversed the order in which the narratives
were presented for two of the lists. All Dutch stimulus materials are available upon
request (see Supplementary Information A1).
3.2.2 Procedure and Data Acquisition
Before the experiment started, participants read and signed an informed consent
form (available from the corresponding author upon request) detailing the nature
of the materials and the procedure as well as emphasising participants’ right to
withdraw consent at any time during the experiment without being required to
provide a reason or losing the right to compensation. Participants were seated in a
comfortable chair in a sound‐proof booth and received verbal instruction. Stimuli
were presented as illustrated in Table 1 above in Times New Roman (font 26) at a
distance of approximately 60 cm. Presentation rate between narratives was self‐
56
timed and two longer pauses were inserted to create three, roughly equal blocks.
Experimental trials were preceded by two practice trials to acquaint the participant
with the procedure. Facial EMG activity was measured continuously using reusable
Ag/AgCl electrodes with a 2 mm contact area over corrugator and zygomaticus
muscles on the right side of the face (van Boxtel, 2010).
To maintain compatibility with similar studies addressing emotional
valence, we included both corrugator and zygomaticus. However, the zygomaticus
does not track emotional valence in the same way (Larsen, Norris, & Cacioppo,
2003). Particularly in more complex situations such as our narrative stimuli, smiling
activity may be difficult to interpret in terms of pure valence. For instance, smiles
can be wry, sarcastic, and smirking as well as expressions of true positive feeling.
We therefore focused on corrugator activity and report the zygomaticus data in
Supplementary Information B2 for reference. Raw EMG signals were recorded with
a NeXus‐10 MKII biosignal system (Mind Media) at a sampling rate of 2048 Hz.
After finishing this part of the experiment, electrodes were removed and
participants moved to a laptop to fill out some questionnaires. Two questionnaires
were included to investigate, in exploratory fashion, potential differences between
individuals in the way simulation and evaluation contribute to corrugator EMG
activity during online language processing: the Adolescent Measure of Empathy
and Sympathy (AMES, Vossen, Piotrowski, & Valkenburg, 2015) and the Moral
Foundations Questionnaire (MFQ, Graham, et al., 2011). Because of the
exploratory, secondary nature of this investigation, we report on the method and
results in Supplementary Information E2.
3.2.3 Data Preparation and Analysis
The raw data were band‐pass filtered between 20‐500 Hz (48 dB/octave roll‐off)
and were additionally filtered with a notch filter at 50 Hz (see van Boxtel 2010),
followed by signal rectification and segmentation per narrative using BrainVision
57
Analyzer 2. For each trial, the 2000 ms of baseline activity preceding the narrative,
consisting of a neutral distractor image of a forest scene, were inspected visually
for remaining artefacts. We selected maximally long epochs of artefact‐free
baseline signal, with a minimum length of 500 ms for both muscles simultaneously.
If such a 500 ms baseline epoch could not be found, the trial was excluded from
analysis (resulting in 1.0% lost trials).
After baseline selection, the data were exported to MatLab for further
segmentation into three parts, time‐locked to the onset of the character morality
segment (5000 ms), the affective state adjective (1000 ms), and the affect reason
(2500 ms). Each of the resulting fEMG segments was then divided into consecutive
100‐ms bins for a balance between good temporal resolution and sufficient
random error reduction. The average facial EMG activity during each bin was
expressed as a percentage of the pre‐narrative baseline activity level (expressing
EMG activity as a percentage of baseline reduces random variance both within and
between individuals; van Boxtel, 2010).
The three critical segments (character morality, affective state adjective,
and affect reason) were analysed separately using the mixed models linear
regression procedure in SPSS (IBM, v24). The model was built iteratively using the ‐
2LL chi‐square test (p <.05), see Supplementary Materials A for a complete report.
In order to be able to evaluate the corrugator response over time we first built a
growth curve model (Mirman, 2015); linear, quadratic, and cubic trends were
added as covariates in the fixed part of the model (trend components were centred
to avoid correlation between trends). By using trends up to the cubic component,
we achieve some flexibility to fit responses without over‐fitting or losing
explanatory power (Mirman, 2015). While models were fitted with a resolution of
100 ms, the parameter estimates (e.g., b for a linear slope) are reported per second
for ease of comprehension. We added separate parameters of time trend
components for each condition to maintain flexibility in building the model and to
58
avoid forcing the model to fit, for example, a quadratic trend for all conditions
when only one condition contained a significant quadratic component.
To assess the effect of our manipulations on the average corrugator
activation over the entire time‐window, we also added character morality as a fixed
factor for the character morality segment. For the affective state adjective and
affect reason segment we added character morality and valence as well as their
interaction as fixed factors. The random part of the model always included random
intercepts for subject and item, as well as random slopes for subjects for each time
trend that initially improved the model.
3.3 Results
3.3.1 Character Morality
Figure 1 shows that reading about moral and immoral behaviours evoked a clearly
differential corrugator response. Immoral actions elicited a rapid increase in
frowning, whereas moral actions did not cause the corrugator response to deviate
from baseline. The statistics corroborated this picture. A main effect of character
morality indicated a significantly higher activation for immoral actions than for
moral actions on the intercept, reflecting the average overall activation7 (difference
b = 74.02, t (261.87) = 16.40, p < .001, 95% CI [65.14, 82.91]). Immoral actions also
revealed a significant linear increase in corrugator activity (b = 29.55, t (60.00) =
4.17, p < .001, 95% CI [15.36, 43.73]), as well as significant quadratic and cubic
components (see Supplementary Materials A). Moral actions, on the other hand,
resulted in no significant increase or decrease in corrugator activity causing the
model to fit a flat line8.
7 Because our growth curve analysis used centred time components, the intercept in our analyses represents the average activation over the entire time window, rather than the point where the curves leave the Y‐axis. Furthermore, whereas the reported intercepts are based on the fixed effects only, the fitted line in Figure 1 includes both the fixed effects and the random effects. 8 The corrugator responses in Fig. 1 start slightly above the 100% baseline level. We suspect this reflects the additional effort demanded by reading the preceding introduction segment of the narratives, in comparison with passive viewing of a preceding neutral image during baseline.
59
We expected a small decrease for moral conditions based on our previous
results ('t Hart B. , Struiksma, van Boxtel, & Van Berkum, under review). That we do
not find this here was not problematic. Many studies have found an asymmetry in
responding to positive and negative valence, with weaker and less reliable
responses to positive stimuli (for a recent discussion, see Alves, Koch, & Unkelbach,
2017). More concretely, our moral actions are probably considered normal
behaviour and therefore limited in how much positive affect they generate,
whereas immoral actions are much more saliently negative. Importantly, the
substantial increase for immoral actions did indicate a clearly negative affective
response, and as such the corrugator did respond differentially to character
morality. This replicates the result of our previous experiment and sets the stage
for the affective event manipulation.
Figure 3.1 Observed averages of corrugator responses to character morality, with growthcurve model regression overlaid.
60
3.3.2 Affective State Adjective
The affective state adjective describes the emotional state of the character (e.g.,
happy vs. frustrated). It is the first word that reveals the valence of the affective
event for the character. In terms of average activation over the entire time window
(the first shaded segment in Figure 2), the model revealed a main effect of valence
(difference negative‐positive b = 6.70, t (252.70) = 2.93, p < .01, 95% CI [2.19,
11.20]), suggesting that character morality did not affect the corrugator activity
patterns. However, the model did reveal some clear time trends that significantly
differed between character morality conditions. Our predictions primarily
concerned the corrugator response to positive and negative events in relation to
character morality. We will therefore discuss the time trends of the moral and
immoral conditions separately below.
Moral Characters. The squares in the first shaded segment in Figure 2 display a
clear differential pattern for positive and negative states ascribed to moral
characters. Negative states evoked an increase in frowning activity while positive
states evoked a decrease. The model corroborated this interpretation (solid lines).
We found a significant linear increase in frowning activity for the moral‐negative
condition, i.e., good people feeling bad (b = 23.25, t (244.52) = 3.02, p < .01, 95% CI
[8.11, 38.39]). In contrast, corrugator activity displayed a significant linear decrease
for the moral‐positive condition, i.e., good people feeling good (b = −11.75, t
(37578.69) = −5.15, p < .001, 95% CI [−16.21, −7.28]). These two linear trends also
differed at the p < .001 level. In addition to a linear trend component, moral‐
negative also contained a significant negative cubic component (see
Supplementary Information C2).
These results are in line with a congruent affective response to positive
and negative affective states ascribed to moral characters. This confirms our
predictions and replicates the findings from our previous experiment. It also shows
61
that it is possible to pick up a clear corrugator response to a single valenced word
in an unfolding narrative.
Immoral Characters. Figure 2 shows a slight decrease in activity for the immoral‐
negative condition and neither an increase nor a decrease in activity during the
immoral‐positive condition (triangles and dotted lines). This pattern also emerged
from the statistical analysis. The corrugator response in the immoral‐negative
condition, that is, bad people feeling bad, revealed a significant linear decrease (b =
−7.18, t (60.64) = −2.25, p = .03, 95% CI [−13.54, −0.81]), while for immoral‐positive,
that is, bad people feeling good, the model fitted a flat line (see Supplementary
Materials A).
3.3.3 Post‐hoc Analysis of Neutral Segment.
The decreasing activation for immoral‐negative during the affective state adjective
was somewhat unexpected, both in light of previous results ('t Hart B. , Struiksma,
van Boxtel, & Van Berkum, under review) and in conjunction with other results in
the experiment at hand (no increase or decrease was found for immoral positive).
To help interpret these results correctly, we performed a post‐hoc analysis of the
neutral segment between the affective state adjective analysis window and the
affect reason analysis window (see Fig 2, 4.75 to 7.25 s). As can be seen in Figure 2,
the predominant pattern during this neutral segment is a surprisingly stable
continuation of corrugator activity levels that were reached at around 1 second
after presentation of the affective state adjective. The analysis of corrugator EMG
responses in this segment revealed no significant difference between the average
activation for immoral‐negative and immoral‐positive (difference b = −0.48, t
(382.27) = −0.10, p = .92, 95% CI [−10.42, 9.45]) while the difference between
moral‐negative and moral‐positive remains (difference b = 19.17, t (382.34) = 3.79,
62
p < .001, 95% CI 9.24, 29.11]). In the face of this highly stable pattern of results, we
are cautious to over‐interpret the modest and short‐lived decrease in corrugator
activity that we observed to negative state adjectives pertaining to immoral
characters, and willing to consider the possibility that it is a potential type I error
induced by random fluctuations in the signal. We will return to this in the
discussion.
Figure 3.2. Observed averages of corrugator responses to the affective event, with the two critical segments (affective state adjective and affect reason) highlighted and vertical lines indicating the onset and offset of other segments (including intersegmental 250 ms intervals).
3.3.4 Affect Reason
At the affect reason segment the reader learned of the circumstances that led to
the character’s affective state, that is, what happened to make the character feel
that way. The second shaded segment in Figure 2 above shows the fitted regression
lines for the affect reason segment. One striking aspect of the results was the
renewed phasic response during the affect reason segment after the relative
stability of corrugator activity patterns after the affective state adjective segment.
63
Our predictions concerned the corrugator response in relation to the valence of the
event, depending on character morality. We will therefore discuss the moral and
immoral conditions separately, starting once again with the conditions containing
moral characters.
Moral Characters. For positive and negative events befalling moral characters we
saw a renewed and sizeable differential pattern emerge: a substantial increase in
frowning for negative affect reasons and a relatively modest decrease for positive
affect reasons (squares in Figure 2). The analysis confirmed this pattern. The model
revealed a significant difference between the average activation for moral‐negative
and moral‐positive conditions (difference negative‐positive b = 41.76, t (265.69) =
8.76, p < .001, 95% CI [32.37, 51.14]). Additionally, the corrugator response to
moral‐negative conditions contained a sizeable linear increase in frowning activity
(b = 18.91, t (60.00) = 2.86, p < .01, 95% CI [5.68, 32.13]), while moral‐positive
conditions showed a modest, but significant linear decrease of corrugator activity
(b = −3.51, t (94530.94) = −3.48, p < .001, 95% CI [−5.49, −1.53]). The difference
between the linear estimates for the two conditions was significant at the p < .001
level. The model also contained a quadratic and cubic component for moral‐
negative, but although these trends initially improved the model, they did not
ultimately prove to be significant (see Supplementary Information C2). These
results conceptually replicated the results from our previous study. Moreover, they
also strongly resemble the differential, phasic response for moral conditions during
the affective state adjective earlier in the sentence, but the size of the corrugator
response is much larger here. We return to this point in the discussion.
Immoral Characters. As for immoral characters (triangles Figure 2), the immoral‐
positive condition elicited a gradual decrease in activity of the corrugator while the
immoral‐negative condition did not evoke a clear decrease or increase. The
difference between the two immoral conditions on average activation overall was
64
not significant (difference negative‐positive b = 7.79, t (250.33) = 1.66, p = .01, 95%
CI [−1.46, 17.03]). The immoral‐positive condition was confirmed to evoke a
modest but significant linear decrease in corrugator activity (b = −2.67, t (64530.94)
= −2.65, p < .01, 95% CI [−4.65, −0.70]). The model confirmed that the best fit for
the immoral‐negative condition was indeed a flat line, as none of the time trends
significantly improved the model (see Supplementary Materials A). Taken together
these results indicated a subtle differential pattern over time for immoral‐positive
and immoral negative.
Switching to a comparison of the two positive conditions, we found that the linear
decrease for immoral‐positive did not differ significantly from moral‐positive
(difference moral‐pos‐immoral‐pos b = −0.83, t (94530.94) = −0.58, p = .559, 95% CI
[−3.63, 1.96]) and neither did the difference in overall average ac va on, although
it did approach significance (difference moral‐pos‐immoral‐pos b = 9.18, t (250.20)
= 1.96, p = .052, 95% CI [−0.07, 18.42]). This suggests that positive affect reasons
elicited a similar corrugator response, regardless of the moral identity of the
character. This constitutes a new finding compared to our previous study and we
will discuss its implications for the multiple‐drivers model below.
3.4 Discussion
This study had two related aims. Firstly, using largely the same materials, we hoped
to replicate findings from a previous experiment ('t Hart B. , Struiksma, van Boxtel,
& Van Berkum, under review); findings that suggested a multiple‐drivers model
(simulation and evaluation) for corrugator activity during online affective language
comprehension. Secondly, we refined the stimulus design to further investigate the
temporal development of the two drivers we proposed in our model: language‐
driven simulation and moral evaluation. Participants read narratives where we
manipulated the moral status of the character and the valence of a subsequent
affective event befalling these same characters.
65
3.4.1 Frowning upon Immoral Behaviour
At the character morality manipulation segment, we predicted a clear effect of
moral valence on the corrugator response. This prediction was confirmed by our
results. Compared to pre‐story baseline, participants frowned more at immoral
actions, but not at moral actions. The presence of a differential effect for
corrugator EMG in response to moral and immoral behavior replicates our earlier
findings and lends further support to the link between corrugator activity and
moral valence, extending the research linking facial EMG to affectively salient
language (e.g., Fino, Menegatti, Avenanti, & Rubini, 2016; Foroni & Semin, 2013;
Havas, Glenberg, Gutowski, Lucarelli, & Davidson, 2010). The successful character
morality manipulation also meant that the stage was set for the reader’s evaluation
of the subsequent critical segments in terms of fairness.
3.4.2 Processing Character Affect
The affective event sentence contained two critical segments with affectively
salient information. At the first, the state adjective, readers discovered how the
protagonist felt about the event. Under the multiple‐drivers model ('t Hart B. ,
Struiksma, van Boxtel, & Van Berkum, under review), we predicted that, for moral
characters, simulation and evaluation would align in terms of valence and thus lead
to a clear differential corrugator response to negative and positive emotions
ascribed to them. This was precisely what we found, corrugator EMG displayed a
clear differential response to good people experiencing negative and positive
emotions. The pattern contained a clear increase for negative states (negative for
the character and evaluated as unfair) and a decrease for positive states (positive
for the character and evaluated as fair).
66
For immoral characters, the multiple‐drivers model predicted that
simulation and evaluation would counteract each other and this would dampen any
differential response to positive and negative state adjectives. Our results revealed
neither an increase nor a decrease for immoral characters experiencing positive
emotions. For immoral characters experiencing a negative emotion, we found a
modest, short‐lived decrease in frowning activity. This pattern would fit with an
explanation in terms of counteracting forces of simulation and evaluation.
However, as already mentioned we hesitate to over‐interpret this particular
decrease and are more confident in the stable absence of a difference between the
two immoral conditions in the following neutral segment. Regardless of whether
the phasic decline in corrugator activity in the immoral‐negative condition should
be taken seriously or not, the overall pattern of results that emerged in response to
the affective state adjective and continued for seconds thereafter – a much
attenuated differential corrugator EMG response to positive and negative events
befalling immoral characters – once again strongly suggests a combined influence
of both evaluation and simulation while processing character affect, rather than
any one single driver controlling the corrugator.
We noted in the introduction that the multiple‐drivers model leaves open
the possibility that simulation and evaluation do not affect the corrugator at the
same time, but might do so successively. We reasoned that the best chances of
finding such a brief ‘simulation‐only phase’ would be at the start of the affective
state adjective segment. Despite a fairly fine‐grained temporal resolution (100 ms)
we found no evidence that simulation emerged before evaluation in corrugator
EMG, but does play a significant role in determining the corrugator activity (as
evidenced by a strong main effect of valence) Based on the evidence here, we
would conclude that, under the multiple‐drivers model, both simulation and
fairness based evaluation influence corrugator EMG at the same time.
67
3.4.3 Processing Reasons for Character Affect
This segment gave the reason for the affective state in the earlier critical segment
by describing the events that evoked them. The affective information was spread
out over a sub‐clause and thus the time‐locking was less precise compared to the
affective state adjective segment. We predicted to conceptually replicate the
findings from our previous study (Larsen, Norris, & Cacioppo, 2003) and find
evidence in support of the multiple‐drivers model. Good and bad things happening
to moral characters evoked a differential corrugator response, showing negative
affect for bad things and positive affect for good things. This replicates the results
from the earlier affective state adjective segment. It also conceptually replicates
the results from the previous study, where the affective event was presented as
one complete sentence.
We predicted that the differential corrugator response to positive and
negative events befalling bad characters would be attenuated compared to good
characters. This was prediction was confirmed. Bad things happening to bad people
resulted in neither an increase nor a decrease in corrugator EMG, while description
of bad characters experiencing a positive event elicited a decrease in activity
(indicative of positive affect). Taken together this resulted in a much reduced
differential corrugator pattern (compared to good character) in response to bad
character experiencing positive and negative events, replication previous findings
and providing further support for the multiple‐drivers model.
It is interesting to note that, on top of the stable pattern elicited by the
affective state adjective, the affect reason segment elicited another sizeable phasic
corrugator EMG activity. This shows that the affect reason segment was in fact a
salient source of additional affective information. Moreover, the amplitude of the
response during the reason segment is much larger than during the affective state
adjective, suggesting that perhaps the evaluation of the affective sentence as a
whole is fully realised when the reason for the affective state of the character is
68
known. The possibility that evaluation of the affective state adjective alone is
incomplete and requires knowledge of the reason could also be further
investigated by manipulating the severity of moral transgressions and the intensity
of the affective events that follow.
3.4.4 Waiting to Update the Situation Model?
The 2500 ms neutral segment following the affective state adjective is intriguing
because corrugator activity looks to be stable for all four conditions. This offers a
tantalising hint as to what manner of language‐driven simulation is involved:
lexical‐conceptual meaning (e.g., ‘frustrated’ or ‘happy’) versus building a situation
model (imagining Mark being furious; cf. Zwaan & Radvansky, 1998; Zwaan &
Kaschak, 2008). Following the multiple‐drivers model, the apparent stability of the
corrugator signal elicited throughout the neutral segment (3 seconds), suggests
that the simulation driver has a persistent effect. Such persistence points to
situation model level simulation, because simulation in service of lexical‐conceptual
retrieval likely be more short‐lived in nature. This would suggest that during this
neutral segment readers are maintaining an active representation of the situation
model they have constructed thus far as they wait to find out the final piece of
affective information.
3.4.5 Open Questions
Although our results support the multiple‐drivers model, we can imagine an
alternative account being put forward where readers selectively simulate affective
meaning depending on whether they have identified with the character or not.
Identification as a concept is defined by Oatley (Oatley, 1999) as “taking on the
character’s goals and plans [as a result of which the audience] experiences
69
emotions when these plans go well or badly.” Several studies have demonstrated a
relationship between character likeability (similar to our character morality
manipulation) and the degree of identification with these characters (Tal‐Or &
Cohen, 2010; Chory, 2013). Other work showed that the level of identification with
a character influenced the degree to which readers reported experiencing
emotions (both positive and negative) in response to the vicissitudes befalling that
character (Hoeken & Sinkeldam, 2014).
While we do not exclude the possibility that identification is a potentially
relevant and interesting factor, we do not see how identification‐mediated,
selective evaluation alone could account for our results. Given that the differential
response to positive and negative state adjectives and affect reasons both included
conditions where corrugator EMG neither increased nor decreased, a selective
simulation account would mean that readers, in those cases, also did not evaluate
what they read at all. Moral evaluation of conspecifics is at the heart of the core
notion of what is fair and unfair. In addition, one important reason for the
pervasive practice of gossip is to share social information and align notions of
moral behaviour (e.g., Dunbar, 2004); this could not work without the continuous
evaluation of good as well as bad characters as the narrative unfolds, particularly if
it concerns moral and immoral behaviour.
3.4.6 Conclusion
Taken together, the results from all three critical segments broadly replicate the
findings from our previous study ('t Hart B. , Struiksma, van Boxtel, & Van Berkum,
under review) and lend support to the multiple‐drivers model. We saw once more
that readers literally frown upon morally objectionable actions described in a story.
The clear dependence of the corrugator response to affective information,
specifically character affect, on the moral status of the character once again
indicates that evaluation does exert control over the corrugator and speaks against
a simulation‐only model for corrugator activity. At the same time, despite a clear
70
role for evaluation, it seems unlikely this is the only driver behind the corrugator
response we observed here. Under the assumption of the salience and dominance
of fairness based evaluation, if the evaluation‐blocks‐simulation model had been
true, the patterns found for moral conditions during the affective state adjective
and affect reason segment should have been reversed for the immoral conditions.
Instead we found plenty of evidence that suggested that simulation also played a
role and we found no such pattern reversal between moral and immoral
conditions. Thanks to our fine‐grained temporal analysis, we achieved more
detailed insight into the online development of simulation and evaluation and
concluded that they most likely exert control over the corrugator concurrently. All
together, these results demonstrated again that the unqualified interpretation of
the corrugator response in terms of language‐driven simulation is untenable. Our
results underscore the importance for a grounded account of language
comprehension of unpacking the complexities of affective meaning in terms of, at
least, evaluation and simulation as we scale up from single words and short
sentences to more ecologically valid language stimuli.
71
4. Study 3
Refining the Multiple‐Drivers Model: Minimal groups vs. Morality
4.1 Introduction
Part of the attraction of stories is the vicarious experience of the lives of the
characters in them. Through stories, we can experience a character’s joy when they
find love, frustration when they quarrel, anguish when they break up, and fury
when they see him or her with a new lover; all the while reclining in our arm chairs
or waiting for the train. According to grounded language processing theories, this
vicarious experience is more than a figure of speech and understanding the words
on the page involves simulation of the events and situations described by them.
Simulation refers to a core notion of grounded cognition: rather than making use of
amodal symbols, conceptual processing depends on multi‐modal representation in
modality‐specific systems; the same systems that support perception, action, and
internal states (e.g. Barsalou, 2008; Anderson, 2010; Barsalou, 2016). Concretely,
this means that understanding the word ‘fury’ involves traces of activity in those
areas of the brain that are active when you are actually furious. Evidence for the
claim that language comprehension involves simulation has accumulated (for
reviews see Zwaan & Madden, 2005; Willems & Casasanto, 2011; Kiefer &
Pulvermüller, 2012), but the field is still developing.
72
Some of the evidence for simulation in language comprehension comes
from studies using facial electromyography (facial EMG). Such studies primarily rely
on the corrugator muscle, which indexes valence with increased activity (frowning)
to negative stimuli and decreased activity (relaxation) to positive stimuli (Larsen,
Norris, & Cacioppo, 2003; van Boxtel, 2010). When language stimuli referring to
positive and negative emotions are processed, the corrugator response has been
found to be similar to the corrugator activity observed when those emotions are
actually experienced (e.g., Foroni & Semin, 2009; Glenberg, Webster, Mouilso,
Havas, & Lindeman, 2009; Havas & Matheson, 2013; Künecke, Sommer, Schacht, &
Palazova, 2015; Fino, Menegatti, Avenanti, & Rubini , 2016). This corrugator activity
during the processing of positive and negative affective language is often
interpreted as a downstream effect of simulation in support of language
comprehension.
However, as we process language, we do not only dispassionately
understand the meaning of words and the situation they describe, we also evaluate
that situation (Van Berkum J. J. A., in press). This evaluation can be crucial for our
understanding of the situation described. For example, imagine reading the same
description of a character’s anger at seeing their ex‐partner with a new lover when
a) their ex‐partner cheated on them with that new lover or b)our character was the
one who cheated and caused the relationship to fail.
In two earlier studies ('t Hart, Struiksma, van Boxtel, & Van Berkum, under
review & in prep.) we manipulated language‐driven simulation and evaluation
orthogonally in short narratives. Each narrative contained a description of a
character behaving either morally or immorally in a given situation. In a subsequent
passage, we manipulated the valence of an affective episode befalling that same
character. In the case of moral characters we found a large differential corrugator
response to positive and negative events befalling them, reflecting positive and
negative affect, respectively. For immoral character on the other hand, we found
no difference between positive and negative events at all in the corrugator
73
response. These results showed that something more than simple lexical or
situation‐model simulation determined corrugator activity during affective
language processing. We also argued that evaluation alone could not be
responsible for this pattern as this would suggest that we do not evaluate valenced
events befalling immoral characters. We consider this explanation highly unlikely,
in part because socially important uses of narrative in human exchanges, such as
gossip, can only work if moral evaluation does not shut down in the case of
immoral or disliked individuals (e.g., Dunbar, 2004). Based on these results we
argued for a multiple‐drivers model of corrugator activity, determined by both
simulation and evaluation. These two drivers align in the case of moral characters
and as such result in a clear differential corrugator response. In the case of immoral
characters, however, these drivers counteract each other and considerably reduce,
eliminate or even reverse the differential corrugator response.
Our previous studies described above both used character morality to
operationalise the reader’s evaluation of the affective episode. Our results thus far
suggest that the counteracting forces of simulation and evaluation eliminate the
differential corrugator response to positive and negative affective information
pertaining to immoral characters. Under the multiple‐drivers model, it should be
possible to systematically manipulate the strength of the evaluation and thus
influence the outcome of the counteracting forces of both drivers. As human
beings, we tend to believe ourselves to be highly moral and virtuous, more so than
the average person (Tappin & McKay, 2016). As such, immoral characters represent
a strong outgroup, which might explain the fact that evaluation and simulation
seemingly cancel each other out. In the study at hand we therefore chose to
contrast the strong morality‐based group distinction with a minimal group
paradigm (e.g., Tajfel, Billig, Bundy, & Flament, 1971; Diehl, 1990). Minimal groups
have already been shown to affect, among other things, how we process words and
faces related to in‐ or outgroup members (e.g., Morrison, Decety, & Molenberghs,
2012; Ratner & Amodio, 2012) and we therefore expect this distinction to also
74
cause simulation and evaluation to counteract each other. However, because
minimal groups are enacted based on comparatively trivial and meaningless
characteristics compared to a moral judgement, the strength of the evaluation
driver for minimal group characters will be comparatively smaller, shifting the
comparative contributions of simulation and evaluation to actual corrugator
activity.
The stimuli in the study at hand contained three critical segments, time‐
locking the corrugator signal to specific affectively salient information: the
character manipulation, the affective state adjective, and the affect reason (see
Figure 1). For the character manipulation segment we predicted that characters
described as moral in‐ or outgroup members (i.e., ‘Mark is a good/bad person’)
should elicit a differential corrugator response, with an increase reflecting negative
affect for immoral characters. For characters described as minimal in‐ or outgroup
members (i.e., ‘Mark is a type P/type O person’) we also predicted a differential
corrugator response, but that this differential pattern should be attenuated
compared to moral in‐ and outgroup because it is the weaker of the two character
manipulations.
Figure 4.1. Stimulus example illustrating the trial structure with the segment names
(top) and presentations times (bottom). Note: between each segment was a fixed
0.25 s delay.
75
During the affective state adjective segment (i.e, ‘happy’ vs. ‘frustrated’)
we predicted to replicate results from our previous studies (notably ‘t Hart et al. in
prep.) for the morality‐based conditions. Specifically, we predicted a clear
differential corrugator response to positive and negative state adjectives attributed
to moral ingroup characters (i.e., ‘Good Mark’). This basic sensitivity of corrugator
activity to valence will be referred to as the differential valence effect and involves
significantly higher activity for negative stimuli than for positive stimuli. For
affective state adjectives describing the state of moral outgroup characters (i.e.,
‘Bad Mark’) we predicted that, because of the counteracting forces of simulation
and evaluation, the differential valence effect should be strongly reduced,
eliminated, or even reversed.
For minimal ingroup characters (i.e., ‘Type P Mark’) we predicted, similar
to moral ingroup characters, a differential valence effect, possibly smaller
compared to moral ingroup‐characters because minimal groups evoke weaker
evaluation. For minimal outgroup characters (i.e., ‘Type O Mark’) we predicted that
the differential valence effect would once again be attenuated, eliminated, or
reversed as a result of counteracting forces of language‐driven simulation and
evaluation. Crucially, however, we predicted that the reduction of the differential
valence effect should be smaller for minimal outgroup members than for moral
outgroup members. Therefore the corrugator patterns in response to characters
experiencing positive and negative emotions should be more similar for minimal in‐
and outgroup characters than for moral in‐ and outgroup characters. Figure 1
below illustrates the predicted patterns for morality‐based and minimal group‐
based conditions in terms of the differential valence effect.
76
Figure 4.2: Illustration of the differential valence effect for morality‐based (A) and
minimal group‐based conditions (B) at the Affective State Adjective Segment.9
During the affect reason segment, the event that elicited the emotion
mentioned in the affective state adjective segment is revealed. The affective
information in this segment was distributed over a multi‐word clause. This meant
the time‐locking of the corrugator signal to affective content was less precise. We
made no detailed predictions for this segment, except that we expected them to on
the whole resemble those observed in previous studies; clear differential valence
effect for moral and minimal ingroup characters experiencing positive and negative
events and some indication of a counteracting drivers in the case of moral and
minimal outgroup characters.
4.2 Method
4.2.1 Participants and Stimulus Materials
We recruited 64 participants (6 male) aged between 18 and 27 (M = 21.48, SD =
2.19) from the Utrecht University Humanities database. All were native speakers of
Dutch, without dyslexia and without Botox® injections to the face. At the time of
testing, our institute did not have an internal review board, but we complied with
standard practices for experiments with human participants. Participants signed
9 Note: Only linear time components included for purpose of illustration, but analysis will include quadratic and cubic components.
77
informed consent forms that detailed the procedure and the nature of the stimuli.
The informed consent form also stressed the voluntary nature of their participation
and their right to withdraw consent at any moment and still receive the stated
financial compensation (€ 12,‐).
We used 64 sentences describing affective episodes taken from larger
narratives used in a previous experiment ('t Hart, Struiksma, van Boxtel, & Van
Berkum, under review). The stimulus sentences were written according to the
structure outlined in Figure 1. Each sentence had eight variants based on our 2 x 2
x 2 design: grouping dimension (morality vs. minimal) x group (ingroup vs.
outgroup) x episode valence (positive vs. negative). Based on this design we made
8 lists of 128 trials that were composed such that a) half started with a 64‐item
block of morality items and the other with a block of minimal group items, b) every
affective episode occurred only once in each block in only one of the four variants
of group x episode valence, c) for each block, 32 affective episodes described a
male protagonist and 32 described a female protagonist (counterbalanced for all
but nine affective episodes due to stereotyped behavioural expectations).
The manipulation of grouping dimension and group was instantiated by
presenting an image before the affective episode. The image consisted of a male or
female silhouette with a badge on their chest that said either ‘good’/’bad’ or ‘type
P’/’type O’. Each image had a caption underneath “X is a good/bad person” or “X is
a type O/type P person”. All stimulus materials (in Dutch) are available upon
request (see Supplementary Information A2).
4.2.2 Procedure and Data Acquisition
After signing an informed consent form, participants were seated in a comfortable
chair in a sound‐proof booth and completed a short questionnaire containing
questions to do with superficial aspects of personality, specifically unrelated to
morality. The test automatically returned a result that, unbeknownst to them,
78
always classified them as a Type P personality. They were then asked to enter their
result and submit the questionnaire for administrative purposes. To enhance the
feeling of group membership, we asked them to wear a badge with a capital P on it,
supposedly in case we lost the questionnaire data.
Next, participants received verbal instruction regarding the rest of the
experiment. Stimulus presentation rate was self‐timed between trials and
automatically timed as described in Figure 1. within trials. Each block was preceded
by two practice trials and there was a pause between blocks where participants
performed an unrelated task. Facial EMG was measured continuously using
reusable Ag/AgCl electrodes with a 2 mm contact surface on the right side of the
face over corrugator and zygomaticus sites. We included both corrugator and
zygomaticus to maintain compatibility with similar studies addressing emotional
valence. However, zygomaticus does not track emotional valence in the same way
that the corrugator does (Larsen, Norris, & Cacioppo, 2003). Particularly in more
complex situations such as our narrative stimuli, smiling activity may be difficult to
interpret in terms of pure valence. For instance, smiles can be wry, sarcastic, and
smirking as well as expressions of true positive feeling. We therefore focused on
corrugator activity and report the zygomaticus data in Supplementary Information
D3 for reference.
Raw EMG signals were recorded with a Nexus MKII biosignal system (Mind
Media) with a 2048 Hz sampling rate. Upon completion of both blocks, the
electrodes were removed and participants moved to a laptop to fill out two
questionnaires: the Adolescent Measure of Empathy and Sympathy (AMES, Vossen,
Piotrowski, & Valkenburg, 2015) and the Moral Foundations Questionnaire (MFQ,
Graham, et al., 2011). These questionnaires were only of exploratory interest and
we therefore report on them separately in Supplementary Materials E2. Finally,
participants were debriefed and received payment.
79
4.2.3 Data Preparation and Analysis
facial EMG data. The raw data were band‐pass filtered between 20‐500 Hz (48
dB/octave roll‐off) and were additionally filtered with a notch filter at 50 Hz (see
van Boxtel 2010), followed by signal rectification and segmentation per trial using
BrainVision Analyzer 2. For each trial the 2000 ms of baseline activity preceding the
narrative, consisting of a neutral distractor image of a forest scene, were inspected
visually for remaining artefacts. Data collection error resulted in the loss of data of
two participants. We selected maximally long epochs of artefact‐free baseline
signal, with a minimum length of 500 ms for both muscles simultaneously. If such a
500 ms baseline epoch could not be found, the trial was excluded from analysis
(resulting in 3.45% lost trials).
After baseline selection, the data were exported to MatLab for further
segmentation into three parts, time‐locked to the onset of the character
manipulation segment (3500 ms), the affective state adjective (1000 ms), and the
affect reason (2500 ms). Each of the resulting facial EMG segments was then
divided into consecutive 100‐ms bins; 100ms bins provide a good balances good
temporal resolution and sufficient random error reduction. The average facial EMG
activity during each bin was expressed as a percentage of the pre‐narrative
baseline activity level (expressing facial EMG activity as a percentage of baseline
reduces random variance both within and between individuals; van Boxtel, 2010).
The three critical segments (character morality, affective state adjective,
and affect reason) were analysed separately using the mixed models linear
regression procedure in SPSS (IBM, v24). The model was built iteratively using the ‐
2LL chi‐square test (p <.05), see Supplementary Materials A3 for a complete report.
In order to be able to evaluate the corrugator response over time we first built a
growth curve model (Mirman, 2015); linear, quadratic, and cubic trends were
added as covariates in the fixed part of the model (trend components were centred
to avoid correlation between trends). By using trends up to the cubic component,
80
we achieve some flexibility to fit responses without over‐fitting or losing
explanatory power (Mirman, 2015). While models were fitted with a resolution of
100 ms, the parameter estimates (e.g., b for a linear slope) are reported per second
for ease of comprehension. We added separate parameters of time trend
components for each condition to maintain flexibility in building the model and to
avoid forcing the model to fit, for example, a quadratic trend for all conditions
when only one condition contained a significant quadratic component.
To assess the effect of our manipulations on the average activation over
the entire time‐window, we also added grouping dimension (morality vs. minimal)
and group (ingroup vs. outgroup), as well as their interaction, to the fixed part of
the model for the character manipulation segment. For the affective state adjective
and affect reason segment we added grouping dimension, group and episode
valence as well as their various two‐way and three‐way interactions as fixed
factors. The random part of the models always included random intercepts for
subject and item, as well as random slopes for subjects for each time trend that
initially improved the model.
4.3 Results
4.3.1 Character Manipulation
The character manipulation contained two variables, each with two levels:
grouping dimension (morality vs. minimal) x group (ingroup vs. outgroup). Figure 2
below displays the results for the four conditions. The type III test of fixed effects
revealed a significant interaction of grouping dimension x group, F(3, 117.46) =
2970.73, p < .001. However, post‐hoc tests with Bonferroni adjustment for multiple
comparisons showed that none of the conditions differed significantly from each
other with regards to average activation over the entire time‐window, see
81
Supplementary Information C3 for a full report.10 Because we were interested in
more than the average activation alone, we included time trend components to
investigate the development of the corrugator response over time. Our primary
interest was in discovering whether moral ingroup vs. moral outgroup and minimal
ingroup vs. minimal outgroup elicited a differential corrugator valence effect. We
will therefore discuss these contrasts separately, starting with the morality
manipulation.
Moral in‐ outgroup. The black lines in Figure 2 below show an increase in frowning
activity in response to characters being labelled as moral outgroup (i.e., a ‘bad
person’), while moral ingroup (i.e., a ‘good person’) characters elicited no change in
corrugator activity. The observed averages suggest a differential response to
descriptions of characters as either moral in‐ and outgroup members, but the
statistics are ambiguous. The linear increase in frowning activity in response to
moral outgroup characters was not significant (ẞ = 2.19, t (62.02) = 1.97, p = .053,
95% CI [−0.03, 4.41]). Addi onally, the model also contained a non‐significant,
negative quadratic trend for the same condition (ẞ = −2.88, t (61.98) = −1.87, p =
.066, 95% CI [−5.96, 0.19]), indica ng a rise/fall pattern. For moral ingroup
characters, on the other hand, the model fitted a flat line, a clear indication of the
absence of any significant increase or decrease in corrugator activity. Taken
together, our morality‐based character manipulation did not elicit a clear
differential response to moral ingroup and moral outgroup members, but the
results are such that we also cannot definitively conclude there is no difference at
all. While at trend‐level, the difference that we found was in the expected direction
with higher corrugator activity for moral outgroup characters than moral ingroup
characters.
10 Because trend variables were centred, an intercept reflects the average corrugator activity across the
entire epoch at hand (based on the fixed effects in the final model), not the level at which corrugator activity intercepts the y‐axis.
82
Figure 4.3. Observed averages for corrugator response at character manipulation
segment, growth curve overlaid.
Minimal in‐ outgroup. The grey lines in Figure 3 show a very similar response to
characters being labelled as either a minimal ingroup (a ‘type P person’) or a
minimal outgroup (a ‘type O person’) member. Our analysis showed that the
corrugator response during both minimal ingroup‐ and outgroup conditions
contained a positive linear component. For minimal outgroup characters this
increase was significant (ẞ = 2.26, t (299.09) = 3.63, p < .001, 95% CI [1.04, 3.48]),
but not for minimal ingroup characters (ẞ = 1.19, t (61.90) = 1.90, p = .062, 95% CI
[−0.06, 2.45]). However, pairwise comparison revealed that the difference between
the linear time trends for minimal in‐ and outgroup characters was not significant
(p = .109). In all, the corrugator response did not show a differential response to
minimal group membership.
83
The absence of such a differential response could point to the minimal
group manipulation having been unsuccessful. However, a two‐tailed, paired‐
samples t‐test showed that, in the exit questionnaire, participants did report
feeling significantly less similar (range= −3 not similar at all, 3 very much similar) to
minimal outgroup characters (M = −1.61) compared to minimal ingroup
characters(M = 0.70): M = −2.31, SD = 1.19, t(61) = −9.06, p < .001. In all, even
though participants reported feeling more similar to minimal ingroup characters,
the minimal group manipulation did not evoke a difference in corrugator activity
between seeing a character described as either a type P (ingroup) or type O
(outgroup) person.
4.3.2 Affective State Adjective
At the affective state adjective our design became a 2 (grouping dimension:
morality/minimal group) x 2 (group: ingroup/outgroup) x 2 (event valence:
positive/negative) design. Type III tests of fixed effects revealed a significant three‐
way‐interaction of these factors, F(7, 276.70) = 232.23, p < .001. Figure 4a below
displays the morality‐based conditions and 4b displays the minimal group‐based
conditions.
84
Figure 4.4a & 4.4b. Observed averages for corrugator response at affective state
adjective segment for morality‐based conditions (3a) and minimal group‐based
conditions (3b), growth curves overlaid. Note that, although split across two figure
here, all eight conditions were analysed in a single model.
Morality in‐ outgroup. The response patterns for moral ingroup characters (solid
lines in figure 4a.) showed a clear differential pattern for positive and negative
state adjectives. Our analysis confirmed that, on average, participants frowned less
when moral‐ ingroup characters (i.e., a ‘good person’) experienced a positive
emotion compared to when they experienced a negative emotion (difference b =
−12.69, t (505.90) = −4.91, p < .001, 95% CI [−17.77, −7.61]). The temporal
development of the corrugator response for the moral ingroup conditions also
differed. The model fitted a flat line for moral ingroup‐positive, but included a
significant linear increase in activation for moral ingroup‐negative (b = 56.39, t
(77.04) = 2.65, p = .010, 95% CI [13.95, 98.83]). The model also included a cubic
component for moral ingroup‐negative; see Supplementary Materials A3 for a full
report. In all, average corrugator activity and temporal development combined
confirmed the presence of a clear differential pattern. As such, these results
replicate results from previous studies, notably ‘t Hart et al in prep.
85
The dotted lines in Figure 4a show a slight increase for both positive and
negative state adjectives attributed to moral outgroup characters (i.e., a ‘bad
person’). This increase proved to be significant in the case of moral outgroup‐
negative (b = 9.11, t (75797.11) = 3.17, p = .002), but not for moral outgroup‐
positive (b = 6.59, t (62.23) = 1.06, p = .294). Pairwise comparison revealed neither
a significant difference in average activation over the entire time‐window
(difference b = −0.69, t (507.31) = −0.27, p = .791, 95% CI [−5.77, 4.40]) nor in the
linear trends for the two moral outgroup conditions (difference b = −2.52, t (91.55)
= −0.37, p = .714, 95% CI [−16.16, 11.12]). The absence of such differences suggests
that, although both conditions described a slight increase, the corrugator showed
no sensitivity to the valence of the affective state adjectives in the moral outgroup
conditions. Taken together with the moral ingroup conditions, this constitutes a full
replication of our second study (‘t Hart et al., in prep.) and aligns with our
predictions regarding the ‘dampening’ of the differential valence effect on
corrugator activity (as illustrated in Figure 2a) for moral outgroup characters
compared to moral ingroup characters.
Minimal Group Manipulation. In figure 4b the negative state adjectives showed a
clear increase, regardless of minimal group membership (black lines). Positive state
adjectives showed neither an increase nor a decrease, both for minimal ingroup‐
positive and minimal outgroup‐positive (grey lines). Our analysis showed that
average activation over the entire time window was significantly higher for minimal
ingroup‐negative than for minimal ingroup‐positive (p < .001). Similarly, outgroup‐
negative elicited significantly higher corrugator activity than outgroup‐positive (p <
.001). Furthermore, there was no difference in average frowning activity between
in‐ and outgroup‐negative and between in‐ and outgroup‐positive (p = .931 and p =
.560 respectively).
The temporal development of positive and negative state adjectives did
not vary as a result of minimal group membership. Minimal ingroup‐negative and
86
minimal outgroup‐negative both contained a significant linear increase in
corrugator activity: minimal ingroup‐negative: b = 43.39, t (118.52) = 3.33, p < .001,
95% CI [17.55, 69.24] and minimal outgroup‐negative b = 39.68, t (150.25) = 3.43, p
< .001, 95% CI [16.80, 62.55]. Pairwise comparison revealed these two did not
differ between in‐ and outgroup (difference b = 3.72, t (998.58) = 0.33, p = .740,
95% CI [−18.28, 25.72]). The model included a nega ve cubic trend for nega ve
affective state adjectives, indicating a fall/rise/fall pattern, but there was once
again no difference between in‐ and outgroup members (p = .672).
We had considered the possibility of the differential valence effect for
minimal ingroup conditions to be smaller overall than for moral ingroup conditions.
Pairwise comparisons showed that there was no significant difference between
moral ingroup‐negative and minimal ingroup‐negative; not on the average
activation over time, the linear trend component, or the cubic trend component
(all p > .317), while both moral‐ and minimal ingroup‐positive were fitted with a flat
line. For a full report of all estimates and comparison see Supplementary
Information C3.
Taken together, the presence of equally differential responses to positive
and negative affective states attributed to both minimal ingroup and minimal
outgroup characters suggests that there was no in fact no reduction of the
differential valence effect. This aligns with our predictions only insofar that we
predicted the reduction of the differential valence effect would be smaller for
minimal group‐based conditions than for morality‐based conditions.
4.3.3 Affect Reason Segment
At the affective reason segment participants read the final part of the sentence, a
sub‐clause describing the event that led to the character experiencing the affective
state described previously. This clause was presented in its entirety and for a fixed
duration of 2500 ms. The model once more revealed a significant three‐way
87
interaction between grouping dimension, group, and valence: F(7, 291.26) = 87.89,
p < .001. Our eight conditions are displayed divided by their grouping dimension in
Figure 4a (morality based) and 4b (minimal group based) below. One striking aspect
of the patterns in figures 4a and 4b is the renewed phasic corrugator response in
addition to that observed previously during the affective state adjective segment.
This suggests that the affect reason once again contained enough affectively salient
information to elicit corrugator activity.
Figure 4.5a & 4.5b. Observed averages for corrugator response at affect reason
segment for morality‐based conditions (4a) and minimal group‐based conditions
(4b), growth‐curves overlaid.
Moral in‐ outgroup Figure 4a showed the expected differential pattern for moral
ingroup characters (solid lines): a clear and substantial increase for negative events
befalling moral ingroup characters and a flat‐line for good things befalling moral
ingroup characters. Positive and negative affect reasons described for moral
outgroup characters (dotted lines) also looked to evoke a differential corrugator
response: an increase in response to negative events befalling moral outgroup
characters and a slight increase in response to positive events.
88
The analysis confirms the presence of a differential corrugator response to
positive and negative affect reasons described for both moral ingroup and moral
outgroup characters. For moral ingroup‐negative (i.e., good people experiencing
something bad) we found a clear linear increase in corrugator activity (b = 41.99, t
(82.81) = 4.31, p < .001, 95% CI [22.62, 61.35]), compared to a flat line fitted for
moral ingroup‐positive (i.e., good people experiencing something good). In
addition, the model also contained significant quadratic and cubic time
components for moral ingroup‐negative (both p < .05, see supplementary
information C3). Finally, the average corrugator activation for moral ingroup‐
negative was significantly higher than that for moral ingroup‐positive (difference b
= −61.34, t (554.54) = −10.67, p < .001, 95% CI [−72.65, −50.04]).
For moral outgroup‐negative (i.e., bad people experiencing something
bad) the analysis revealed a significant linear increase (b = 38.77, t (80.34) = 3.79, p
< .001, 95% CI [18.39, 59.15]) and for moral outgroup‐positive (i.e., bad people
experiencing something good) a non‐significant linear increase (b = 4.27, t (61.74) =
1.01, p = .318, 95% CI [−4.21, 12.76]). The difference between these two linear
estimates was significant (difference b = 34.50, t (91.44) = 4.10, p < .001, 95% CI
[17.60, 50.74]). The model also included significant quadratic components for both
moral outgroup conditions and a significant cubic component for moral outgroup‐
negative (all p < .01, see supplementary information C3). Finally, the average
corrugator activity was also significantly higher for moral‐outgroup negative than
moral outgroup‐positive (difference b = 21.90, t (608.66) = 3.72, p < .001, 95% CI
[33.47, 10.33]).
In all, the results above confirmed a differential corrugator response to
positive and negative affect reasons for both moral ingroup and moral outgroup
characters. However, the differential response for moral outgroup conditions
seemed reduced compared to the moral ingroup conditions. The analysis
supported this interpretation with a significant difference on average activation
between the two positive conditions (difference b = 18.44, t (554.55) = 3.21, p <
89
.001, 95% CI [29.75, 7.14]) and a significant negative quadratic (rise/fall)
component for moral outgroup‐positive (b = −6.23, t (190585.27) = −2.60, p = .009,
95% CI [−10.94, −1.53]), compared to a flat line for moral ingroup‐positive. The two
negative conditions also differed in average activation (difference b = 20.99, t
(608.66) = 3.56, p < .001, 95% CI [9.42, 32.56]), though not on either linear,
quadratic, cubic components (all p > .3, see supplementary information C3). Taken
together this suggests that, as predicted, the differential valence effect was
somewhat reduced for moral outgroup conditions, constituting a conceptual
replication of our previous study (‘t Hart et al. in prep.).
Minimal Group Manipulation. Figure 4b showed differential corrugator responses
to positive and negative affect reasons for both minimal ingroup characters (i.e.,
type P people) and minimal outgroup characters (i.e., type O people). The minimal
ingroup‐negative condition evoked significantly more corrugator activity over the
entire time window than did minimal ingroup‐positive (b = 50.59, t (553.83) = 8.79,
p < .001, 95% CI [39.28, 61.89]). Similarly, minimal outgroup‐negative also evoked
significantly more negative affect than minimal outgroup‐positive (b = 43.85, t
(555.90) = 7.61, p < .001, 95% CI [32.54, 55.16]). As for the development of these
responses over time, while both minimal ingroup‐positive and minimal outgroup‐
positive elicited no increase or decrease in corrugator activity, both minimal
ingroup‐negative and minimal outgroup negative displayed significant linear
increases in corrugator activity (both p < .01, see Supplementary Information C3 for
a detailed report).
The comparison between the positive affect reason conditions revealed no
difference in the corrugator response depending on whether something good
happened to a minimal ingroup or a minimal outgroup character (grey lines). Both
were fitted with a flat line and there was no difference on the intercept (b = 0.86, t
(503.24) = 0.152, p = .879, 95% CI [−10.18, 11.89]). Nega ve affect reasons did
prove sensitive to minimal in‐ or outgroup member status. The model revealed that
90
average activation over the entire time‐window for minimal ingroup‐negative and
minimal outgroup‐negative conditions did not differ significantly (b = 7.59, t
(609.00) = 1.29, p = .198, 95% CI [−3.98, 19.16]), but the temporal development of
the two negative affect reason conditions did show a difference. The linear
component for minimal outgroup‐negative was significantly smaller compared to
that for minimal ingroup‐negative (b = −26.07, t (124.78) = −4.07, p < .001, 95% CI
[−38.76, −13.38]). Taken together these results once more suggest a dampening of
the differential valence effect for affective language pertaining to outgroup
characters, in this case minimal outgroup characters.
4.4 Discussion
This study was designed with two aims in mind. Firstly, we aimed to replicate
findings from previous studies (‘t Hart et al. under review & in prep.); findings that
suggested a multiple‐drivers model (simulation and evaluation) for corrugator
activity during affective language processing. Secondly, and more importantly, we
aimed to show that it was possible to systematically manipulate the strength of the
evaluation driver in relation to simulation and as such reduce the differential
valence effect on corrugator activity we found previously. Doing so would provide
further support for the validity of the multiple‐drivers model. For the replication we
once more made use of a character manipulation based on morality, followed by
an affective episode befalling those same characters. To manipulate the strength of
the evaluation we contrasted the effect of the morality‐based character
manipulation with that of a manipulation based on minimal groups.
4.4.1 Morality vs. Minimal Group Manipulation
As predicted, the morality‐based character manipulation evoked a weak differential
corrugator response with slightly higher activation for descriptions of characters as
91
bad as opposed to good. However, this difference proved only marginally
significant. Contrary to our predictions, the minimal group‐based manipulation did
not evoke any differential corrugator response, despite the exit questionnaire
showing that participants did report feeling more alike to minimal ingroup‐ than
minimal outgroup members. The absence of a differential response to the minimal
groups manipulation compared to the, albeit trend‐level, differential response to
morality does align with our prediction that that the minimal group manipulation
would evoke a weaker evaluative response compared to the morality‐based
manipulation.
In previous studies we did find large and robust differential corrugator
responses to morality‐based character manipulations (‘t Hart et al., under review &
in prep.). However in those previous studies, character’s moral status was
manipulated based on a narrative description of the behaviour by characters in a
concrete situation, rather than with an explicit propositional statement. The
absence of a significant differential response to moral status in this study attests to
the power of concrete events in narratives to evoke affective responses compared
to the simple characterisation of story characters with canonically valenced
descriptors such as “good” or “bad”.
4.4.2 Processing Character Affect
Although the effects observed during the character manipulation segment for
morality‐based manipulations were only marginally significant, our predictions
regarding the role character morality would play during the affective state
adjective were confirmed. As predicted we found that, in the case of ‘good’
characters experiencing positive and negative emotions, the corrugator responded
differentially and with a higher level of activation for negative affective states
compared to positive states. For ‘bad’ characters, we predicted that this differential
valence effect would be reduced, eliminated, or reversed. Our results confirmed
92
this prediction as we found no significant difference between corrugator activity
when participants read about ‘bad’ characters experiencing positive or negative
emotions. These results are a direct replication of ‘t Hart et al. (in prep.) and once
more lend support to an explanation of corrugator activity during affective
language processing in terms of the multiple‐drivers model. According to the
multiple‐drivers model, language‐driven simulation and evaluation align for ‘good’
characters, but represent opposing forces in the case of ‘bad’ characters. The
minimal‐group based manipulation was intended to provide a weaker evaluation
compared to the morality‐based manipulation.
As predicted, moral ingroup characters experiencing positive and negative
emotions once more displayed a clear differential corrugator response, with
significantly higher activation for negative affective state adjectives. For minimal
outgroup characters we found an equally differential corrugator response to
positive and negative state adjectives, once again with significantly higher
activation for negative state adjectives. This constitutes an extreme variant of our
prediction that the reduction of the differential valence effect would be smaller for
minimal outgroup characters compared to moral outgroup characters; that is to
say, rather than a smaller reduction, the differential valence effect is unaffected.
This suggests that we evaluate minimal‐ in and outgroups characters experiencing
positive or negative emotions the same.
4.4.3 Processing Reasons for Character Affect
As predicted, we found a differential valence effect on corrugator activity for ‘good’
characters. We also found a differential valence effect for events befalling ‘bad’
characters. Also as predicted,we found evidence of a reduction of the differential
valence effect for ‘bad’ character, that is to say a ‘dampening’ of the differential
response compared to ‘good’ characters. This constitutes conceptual replication of
our previous study (‘t Hart et al. in prep.b), where we also found a differential
93
valence effect for both ‘good’ and ‘bad’ characters during the affect reason
segment. The differential valence effect for ‘bad’ characters in the study at hand,
however, seems much larger than what we found previously. The difference
between these two studies might be caused by the weaker evaluative strength of
the propositional manner in which character morality was manipulated in the study
at hand, as opposed to the concrete event description in the previous study.
Turning to the minimal group conditions, we found the predicted
differential valence effect for minimal ingroup characters. Minimal outgroup
characters experiencing positive and negative affect reasons also once more
evoked a differential corrugator response. Surprisingly, we found the first evidence
of any effect of minimal group membership on the corrugator response. Despite
not differing on average activation, the linear increase in corrugator activity for
minimal ingroup characters experiencing a negative affect reason was significantly
larger than for minimal outgroup negative affect reasons. We found no difference
for positive events befalling minimal in‐ or outgroup members. These results
provide tentative support for a reduced differential valence effect for minimal
outgroup characters. While we would first like to see such a result replicated, this
suggests that a relatively ‘empty’, minimal group distinction is sufficient basis for
differential evaluation during online language comprehension and suggests that
participants cared less about bad things happening to outgroup characters. This is
in line with studies reporting reduced empathy for outgroup characters
experiencing negative emotions (e.g., Cikara, Bruneau, & Saxe, 2012; Montalan,
Lelard, Godefroy, & Mouras, 2012) .
That no such difference was found for morality‐based conditions during
the affective reason segment suggests that the influence of evaluation for minimal
in/outgroup characters and moral in/outgroup characters is perhaps qualitatively
different rather than quantitatively. That is to say, perhaps it is not the strength of
the evaluative driver that differs, but rather what is evaluated. Based on these
results alone, it seems moral group identity influences evaluation of language
94
describing emotions more, while minimal group identity influenced the evaluation
language describing events befalling those characters.
4.4.4 General Discussion
In all, the results support the multi‐drivers model for corrugator activity during
affective language comprehension. A simulation‐only model has once more been
proven to be incorrect as it should have resulted in the same pattern of corrugator
activity in response to characters experiencing positive and negative emotions,
regardless of their moral group status. An evaluation‐block‐simulation model,
where evaluation alone determines corrugator activity, is also unlikely. It has been
quite well‐established that we experience Schadenfreude in response to disliked or
considered undeserving characters experiencing something negative (e.g., Feather
& Nairn, 2005; Leach & Spears, 2009; Singer, et al., 2006) and this has also been
shown to influence corrugator activity (Cikara & Fiske, Stereotypes and
Schadenfreude: Affective and physiological markers of pleasure at outgroup
misfortunes, 2012). Similarly, we might expect a kind of outraged sense of justice
to evoke increased corrugator activity when ‘bad’ characters experience something
good. In all, the most parsimonious explanation of the elimination of the
differential valence effect for moral outgroup characters remains the multiple‐
drivers model. This model helps guide research exploring the role of simulation in
language comprehension and online affective responding, using more natural
language stimuli that engage the kind of complexity our narratives engaged
This study extends previous results by showing that the conflict between
evaluation and simulation can also be enacted by a propositional manipulation of a
character, rather than through concrete event descriptions in a narrative format.
Additionally, it shows that an affective response to the character manipulation is
not necessary to create the opposing forces of the evaluation and simulation
drivers during subsequent affectively salient segments. The comparison of results
95
from the morality‐based and the minimal group‐based dimension shows that the
effects of moral identity are not reducible to in‐ and outgroup status only. At the
same time, we saw that a relatively meaningless grouping dimension such as
minimal groups is sufficient basis for a differential evaluation during online
language comprehension
Taken together, these results once more highlight the importance of
considering the complexity of affective meaning in language processing research
and the dynamics of online affective responding. This is especially important with
regards to the interpretation of corrugator activity as reflecting neural simulation
within a grounded language comprehension framework. At the same time, social
psychologists using facial EMG or similar psychophysiological measures linguistic
vignettes to investigate psychological processes must take into account that
language itself also influences. Although this study has focussed on corrugator EMG
recordings, our findings also call for consideration of this type of affective
complexity in the interpretation of EMG recordings of other affect‐related facial
muscles (e.g., zygomaticus major, orbicularis oculi, frontalis), psychophysiological
measures of emotion (e.g., heartrate and galvanic skin response), but also
neuroimaging studies.
96
97
5. General Discussion
The overall aim in this dissertation has been to shed light on the influence of
processes of evaluation and language‐driven simulation during affective language
comprehension. The introduction provided a framework for this question by pulling
together theories of grounded language processing and emotion. The introduction
also outlined how facial EMG and narrative discourse could be used to study the
role of evaluation and simulation during affective language comprehension. The
next three chapters reported on studies that incrementally provided answers to the
central research question while at the same functioning as (conceptual)
replications.
In this final chapter, the evidence from all three studies will be pulled
together in a discussion of the three models formulated based on the general
processing framework for EMG activity presented in figure 1 in the Introduction.
Moving beyond a discussion in terms of the focussed question posed in this
dissertation, the conclusion will also discuss the results more broadly, including the
character manipulation, other potential relevant factors and alternative
explanations. Throughout, the theoretical implications and possibilities for future
research will be discussed.
5.1 Facial EMG during online affective language comprehension
The introduction presented a framework for corrugator EMG activity during
affective language processing (see Figure 1) and briefly outlined three models of
how language‐driven simulation and evaluation might contribute to that activity:
simulation‐only, evaluation‐blocks‐simulation, multiple‐drivers. The studies
reported in this dissertation were all designed to tests the validity of these three
models.
98
5.1.1 Simulation‐only model
The simulation‐only model described a situation where, at least in lab studies such
as these, language‐driven simulation alone would determine corrugator activity
during affective language processing. Based on this model no difference in
corrugator activity was expected in relation to whether a character was either good
or bad or part of a minimal ingroup or outgroup. This model was considered to be
unlikely, but was included because it reflects the way corrugator activity—as well
as that of other facial muscles—has been interpreted in affective language
processing studies within a grounded cognition framework (e.g., Fino, Menegatti,
Avenanti, & Rubini, 2016; Künecke, Sommer, Schacht, & Palazova, 2015; Havas &
Matheson, 2013; Foroni & Semin, 2013; Havas, Glenberg, Gutowski, Lucarelli, &
Davidson, 2010; Foroni & Semin, 2009; Glenberg, Webster, Mouilso, Havas, &
Lindeman, 2009; Niedenthal, Winkielman, Mondillon, & Vermeulen, 2009). None of
these studies have, to my knowledge, considered the potential conflict between
language‐driven simulation and evaluation. Figure 5.1 below illustrates the way
corrugator activity would arise according to the simulation‐only model.
99
Figure 5.1 Simulation‐only Model
If language‐driven simulation, either as part of the retrieval of lexical concepts (S1)
or as part of the construction of a situation model (S2), alone determined
corrugator activity, the patterns of activity in response to characters experiencing
positive or negative emotions or events would be the same regardless of moral
status or minimal group membership. In other words, a character described as
‘happy’ would always evoke less corrugator activity compared to a character
described as ‘frustrated’, regardless of whether the character is good or bad, an
ingroup member or an outgroup member.
The results from study 1 revealed that character morality did in fact
influence corrugator activity when reading about a character experiencing a
positive or negative emotion in relation to a valenced event (see figure 2.2). For
moral characters, positive and negative affective events evoked the predicted
differential corrugator response. In the case of immoral characters, however, this
100
differential pattern was not found; positive and negative events befalling immoral
characters elicited similar levels of corrugator activity and displayed no real change
in corrugator activity over time. The implication is that evaluation does in fact
influence corrugator activity during real‐time language comprehension and that in
experiments such as these, corrugator activity cannot be taken to reflect language‐
driven simulation alone.
The drawback in study 1 was that the affective event was presented as one
sentence and thus there was no fine‐grained control over when exactly participants
read affectively salient words or phrases. In studies 2 and 3 the design was refined
to time‐lock the corrugator signal to two specific affective segments. The first of
these was the affective state adjective. This segment described the valenced
affective state of the character with a single word (e.g., ‘frustrated’ vs. ‘happy’),
before any additional information that might influence the evaluation was
revealed. As such, this segment offered the cleanest point to measure the
interaction of simulation and evaluation. As in study 1, studies 2 and 3 (figures 3.2
and 4.4) revealed a clear differential response for moral characters (higher
activation for negative‐ than positive affective states), but no differential
corrugator response for immoral character during the affective state adjective. This
corroborated the findings in study 1 and provided support for rejection of the
simulation‐only model. That is to say, even in lab experiments such as these the
evaluation of an affective linguistic stimulus has an immediate effect on corrugator
activity. The implication is that the results of many of the studies cited at the
beginning of this section should perhaps be reconsidered in terms of evaluation as
well as simulation. Explaining corrugator activity in experiments such as these in
terms of language‐driven simulation only is insufficient. This insufficiency becomes
especially salient when scaling up the stimulus materials to engage with the type of
natural complexity this thesis illustrated with the example from Nabokov’s Lolita in
the introduction: would we really simply simulate a paedophile’s happy feelings of
love?
101
5.1.2 Evaluation‐blocks‐simulation model
The evaluation‐blocks‐simulation model held that, in line with the role of narrative
in moral evaluation (e.g. gossip, see Dunbar, 2004) and the automatic nature of
such evaluation (Greene, 2014), corrugator activity in response to reading about
positive and negative emotions of various characters would be determined entirely
by the evaluation of that information, presumably dominated by fairness. While
evaluation is obviously not restricted to fairness, the stimuli used in these studies
were designed to foreground this salient moral dimension. The evaluation‐blocks‐
simulation model assumed that, although simulation of somebody else’s emotion,
or of the lexical concept involved, may surface in corrugator activity in some
controlled and unambiguous situations, in more complex situations online
corrugator activity would be captured by the affective responses related to
evaluation of the situation in terms of fairness.
Specifically, both situations in which a bad character experiences a
negative emotion as well as those where a good character experiences a positive
emotion or event would be evaluated as fair and would elicit positive corrugator
activity. In the latter case, this positive corrugator activity will likely stem from
affective resonance or empathy. In the latter case however, this evaluation as fair
would elicit emotions such as Schadenfreude. These expectations are based on
existing evidence for Schadenfreude causing reduced empathy for outgroup
character (e.g., Singer, et al., 2006) and facial EMG evidence of the experience of
positive affect in response to certain outgroup members experiencing misfortune
(Cikara & Fiske, 2012)11.
11 Note: the study in question did not include corrugator supercilii activity, but only zygomaticus major
activity and found evidence of Schadenfreude effects for envied outgroups, but not for pitied or disgusting outgroups.
102
Similarly, both situations in which a bad character experiences a positive
emotion or event and a good character experiences a negative emotion or event
would be evaluated as unfair and as such would elicit corrugator activity indicative
of negative affect. Figure 5.2 below illustrates this driver model where both S1 and
S2 are blocked and the evaluation driver alone feeds into motor control systems.
Figure 5.2 Evaluation‐blocks‐simulation Model
None of the three studies revealed the predicted reversed differential corrugator
pattern for immoral characters, not for the affective event segment in study 1 or
for either of the two affective segments in studies 2 and 3. As such, this does not
support a processing model where only fairness‐based evaluation contributes to
corrugator activity.
In previous chapters I argued that the many instances where there was no
differential valence effect for immoral characters at all provided strong evidence
against an evaluation‐blocks‐simulation model. This argument hinged on the
103
observation that the only way evaluation alone could account for the absence of a
differential valence effect would be by assuming that readers only evaluate what
happens to moral characters and do not evaluate affective information regarding
immoral characters at all. Again, considering the social importance of gossiping, in
part for the monitoring and regulation of immoral behaviour, it seems highly
unlikely that we do not evaluate affective information pertaining to immoral
characters, if anything, these should be of particular interest to us as humans. That
we are interested in such cases seems intuitively true given our fascination with
immoral characters in stories. Furthermore, the affect reason segments in studies 2
and 3 did reveal a differential corrugator response to positive and negative
affectively salient information for both moral and immoral character (figures 3.2 &
4.5).
However, there is another explanation that cannot be ruled out entirely. While
we based our reasoning on fairness‐based evaluation depending on the group a
character belongs to, it is not unreasonable to assume that other kinds of
evaluation play a role and that these may counteract each other; for example,
automatic empathy with a bad character experiencing something bad combined
with Schadenfreude. This possibility will be discussed further following the
discussion of the third and final model proposed in the introduction.
5.1.3 Multiple‐drivers model
The third and final model considered beforehand was the multiple‐drivers model,
wherein language‐driven simulation and evaluation would determine corrugator
activity together. The multiple‐drivers model offers an explanation for the absence
of a differential pattern of corrugator activity for immoral characters during the
affective events segment in study 1. For example, under a multiple‐drivers model
the corrugator activity increase due to simulation of a frustrated character and/or
retrieving the associated lexical semantics could be levelled out by the corrugator
104
activity decrease associated with the positive emotions (i.e., Schadenfreude or a
feel‐good sense of justice) elicited by the evaluation of this particular situation as
fair when the character in question previously behaved immorally. Similarly, an
immoral character experiencing happiness might elicit positive language‐driven
simulation on the one hand and negative evaluation on the other hand.
Studies 2 and 3 were designed in part to further test the validity of the
multiple‐drivers model and to explore whether there might be traces of pure
simulation before evaluation. This possibility seems pertinent at an intuitive level
where you might expect that simulation in service of understanding would need to
occur before evaluation can take place. The refined time‐locking of the corrugator
signal to specific affective information was useful in answering this question. The
fact that in both studies 2 and 3 we found no trace of a differential corrugator
response to affective states ascribed to immoral characters provided support for a
truly simultaneous multiple‐drivers model, specifically one wherein corrugator
activity is determined by both language‐driven simulation and evaluation from the
start. It remains possible, however, that a very brief simulation phase has gone
unnoticed because of the choice to follow established optimal signal processing
practices (van Boxtel, 2010) that limit temporal specificity to 100 ms chunks.
105
Figure 5.3 Multiple‐drivers Model
Study 3 aimed to provide further support for the idea of multiple drivers by
contrasting a morality‐based character manipulation with one based on a minimal
group paradigm. The minimal group paradigm was developed in social psychology
to create in‐ and outgroups based on a fairly meaningless distinction (e.g., a
numerical over‐ or under‐estimator). The reasoning behind the choice to include
this paradigm here was that immoral characters could be considered to represent
an extreme outgroup. As such, this had a strong impact on the evaluation of
subsequent language describing such characters experiencing positive and negative
emotions and events. Minimal outgroups are determined on relatively trivial
grounds and thus constituted a much weaker outgroup. We predicted that minimal
outgroups would therefore elicit less group‐dependent evaluation. By manipulating
the amount of group‐dependent evaluation, we hoped to influence the outcome of
the supposed counteracting forces of simulation and evaluation. In case of the
106
outgroup, the evaluation component counteracting simulation would be smaller
compared to immoral characters given that the group‐dependent evaluation would
be weaker and thus more simulation should shine through (see figure 4.2).
At the affective state adjective (see figure 4.4), we saw no differential
corrugator response for positive and negative states describing the emotions of
immoral characters. This suggested that once again simulation and evaluation
cancelled each other out. For minimal group conditions we saw the same
differential corrugator response to positive and negative affective state adjectives
for both ingroup and outgroup characters. This meant that the down‐regulation of
group‐dependent evaluation was successful to such an extent that it disappeared
altogether.
At the affect reason segment (see figure 4.5) we saw an attenuated differential
valence effect for outgroup members in both morality‐based and minimal group‐
based conditions. This once more suggested a combined effect of simulation and
evaluation and supports the multiple‐drivers model. The fact that this attenuation
was stronger for the moral outgroup conditions than for minimal outgroup
members also supports the idea that we can systematically manipulate the amount
of group‐dependent evaluation compared to simulation and influence the outcome
of the multiple‐drivers model.
In all, out of all the models considered a priori, the most parsimonious account
of the results presented in this dissertation is the multiple‐drivers model. Exactly
how and where language‐driven simulation and evaluation come together is
beyond the scope of this thesis, but this is an important next question that the field
will have to answer in pursuit of a grounded account of language comprehension.
107
5.2 Complex evaluation explanation
In the discussion of the evaluation‐blocks‐simulation model above, I mentioned
that there was an alternative explanation that could not be ruled out based on the
evidence presented here. This alternative explanation concerns the possibility that
the absence or attenuation of a differential corrugator response for immoral and
outgroup character is the outcome of a combination of group‐dependent
evaluation based on fairness and more generic types of evaluation. This presents
an alternative to the explanation of the corrugator activity patterns as the outcome
of group‐dependent evaluation (fairness) and language‐driven simulation, as
argued for in the multiple‐drivers model.
The absence or attenuation of the differential valence effect in the case of
immoral characters could possibly be the result of a combination of group‐
independent affective empathy/empathic concern and group‐dependent
evaluation based on fairness. Consider the case of an immoral character being
described as ‘angry’ or ‘frustrated’ because of some negative event. We might
predict a corrugator response reflecting negative affect through a process of
automatic affective empathy or concern, regardless of group status.
Simultaneously, group‐dependent, fairness‐based evaluation might evoke a
positive corrugator response. Both types of evaluation are plausible and could,
together offer an alternative explanation to the patterns of corrugator activity as
found in studies 1‐3, without the need for language‐driven simulation. Following
this reasoning, the patterns of activity observed in the moral conditions, which
resemble canonical simulation patterns, might then simply be the result of generic
evaluation. This also opens the door to the possibility that the clear differential
valence effect on corrugator activity as simulation found in many of the studies
above, also reflects this generic evaluation. This might be likely given the previously
stated care these studies took to make sure the single words, or short sentence
they used as stimuli were unambiguously positive or negative.
108
One potential line of evidence that argues against such an explanation is the
literature on reduced or even reversed empathy responses for outgroup‐members.
(e.g., Luo, et al., 2015; Cikara M. , Bruneau, Van Bavel, & Saxe, 2014; Cikara,
Bruneau, & Saxe, 2012; Montalan, Lelard, Godefroy, & Mouras, 2012). These
studies suggest that group‐dependent evaluation has a clear effect on group‐
independent evaluation. Nonetheless, it remains a possibility that the effects we
see are the outcome of a complex process involving an unknown combination of
different types of evaluation.
5.3 Simulation: lexical concepts or situation models?
As discussed before, the research presented in this thesis was not designed to
make any meaningful distinction between the extent to which the corrugator can
reflect simulation of either lexical concepts or simulation of situation models built
up as part of larger discourse comprehension (i.e., ‘angry’ vs. ‘Mark is angry’).
Nonetheless, the remarkably stable patterns of corrugator activity during the
neutral segments separating the two affectively salient segments in study 2 and 3
seem to be able to shed some light on this distinction. The neutral segment in both
study 2 and 3 separated the affectively salient segments from each other and
contained no new affective information: “Mark was angry / when after a few
minutes / he runs out of petrol. The level of corrugator activity evoked during the
affective state adjective remained stable throughout the neutral segment (see
figure 3.2 and appendix 3D). During the subsequent affect reason segment the
affective information contained therein evoked yet another phasic corrugator
response on top of the stable pattern evoked during the affective state adjective.
This stability in the corrugator signal found in both studies, in combination
with the arguments above for a multiple‐drivers model, suggests that what we are
seeing here is at least partly reflective of the incremental updating of a situation
model (in addition to the stable evaluation thereof). If the simulation component of
109
the corrugator signal had consisted only of retrieval of lexical concepts, this would
have arguably evoked a more short‐lived simulation and we would not have found
such a stable pattern. Further research would be needed to confirm this, but the
implication would be that at least the S2 driver in our model is relevant, possibly in
combination with S1. One way to tease the two possible sources of simulation
apart would be to create a context where the lexical simulation and the situation
model simulation do not align.
5.4 Simulation and language comprehension
In this thesis I set out to answer the question of how processes of language‐driven
simulation and evaluation would interact in more ecologically valid contexts where
affective meaning was not as straightforward as in most affective language
processing studies. The outcome of this interaction was investigated using
corrugator EMG. This measure has commonly been used as both an indicator of
simulation in affective language processing research and as an indicator of
evaluation in (social) psychology research. One clear observation that follows from
the combined results of studies 1, 2, and 3 is that corrugator EMG recordings
cannot unequivocally be taken to reflect language‐driven simulation in the type of
complex, yet ecologically valid, contexts created in the stimulus materials. The
results presented here show the clear dependence of the online corrugator
response on the moral identity of the character to which the stimuli in question
pertained. Importantly, this dependence was replicated and refined across three
studies. As such, the studies reported here offer clear evidence against a
simulation‐only interpretation of corrugator activity during language processing
and for the role evaluation plays during real‐time language processing. I have
therefore argued for a multiple‐drivers model that includes language‐driven
simulation and evaluation. However, while the evidence for the presence of
evaluation in the corrugator signal is clear, there is no direct evidence that
110
simulation impacts corrugator activity. As mentioned before, the evidence for
corrugator, or other facial muscles, as indicators of language‐driven simulation
comes from studies where stimuli were unambiguously positive or negative. The
facial EMG effects such studies reported could simply reflect default evaluation
rather than simulation. Future research would have to find a way to determine
whether simulation directly influences corrugator activity.
One way researchers have attempted to show that facial muscle activity
reflects simulation is by investigating a functional or causal connection between
facial muscle activity and comprehension. One such study (Havas, Glenberg, &
Rinck, 2007) reported that inhibition or facilitation of corrugator and zygomaticus
activity affected the speed of processing of valenced sentences, but not positively
valenced ones. Another study found similar results using Botox injections to inhibit
corrugator activity (Havas, Glenberg, Gutowski, Lucarelli, & Davidson, 2010).
However, such effects on comprehension are acsribed to process of afferent facial
feedback, whereby a peripheral process facilitates central processing. The notion of
such a feedback loop does not speak to whether the initial muscle activation
reflects simulation or not. Indeed, Havas, Glenberg, and Rinck (2007) explicitly
report that manipulating facial muscle activity did not affect initial lexial access and
Glenberg et al. (2010) suggest that facial feedback contributes to comprehension of
subsequent language stimuli by effecting central and autonomic changes in
preparation for likely upcoming stimuli. Once again, even if the periphery could be
definitely proven to be functionally related to comprehension in this sense, it still
would not necessarily mean that facial muscle activity actually reflects neural
simulation.
As briefly mentioned in section 1.1.4, the status of simulation as a causally
contributing component of comprehension is still hotly debated (for recent
discussion see Leshinskaya & Caramazza, 2016; Barsalou, 2016). The uncertainty
regarding the status of language‐driven simulation relates to the ongoing debate
on the validity of a grounded account of language processing in particular and
111
cognition in general. One possibility is that simulation as a constituent process of
language processing is much more context‐dependent than much of the extant
research allows. Perhaps recent experiences (i.e., learning about a character’s
moral nature) places forward constraints on what and how subsequent, relevant
affective information (i.e., that same character being ‘angry’) is simulated (for
recent discussions see Yee & Thompson‐Schill, 2016 and Zwaan, 2016). Simulation
conceived of in this manner (Figure 5.5) might also further help tease apart
lexical/conceptual simulation and situation‐model level simulation as the notion of
context‐dependent, forward constraints on simulation imply the incremental
construction of situation‐models. In a sense, such a conceptualisaiton of language‐
drive simulation is by definition a multiple‐drivers model, as it incorporates
evaluation as a parameter for language‐driven simulation.
Future research might consider varying the intensity of the positive and
negative affective states and affect reasons while keeping the level of the moral
transgression constant to provide insight into whether evaluation of the character
places a general forward constraint on language‐driven simulation activity or if
evaluation plays a role in real‐time processing of the relevant affective segments.
For example, someone who litters in a public park and subsequently either
experiences anger because their phone battery dies or because their car has been
vandalized. If evaluation of the character enacts a general forward constraint on
the simulation of subsequent affective information pertaining to that character, the
severity of the affective events that befall this character should make no difference
to the corrugator pattern. I would predict that it does make a difference and that
the differential valence effect will be much more strongly attenuated at the affect
reason segment for less serious events than for serious events. An alternative
version of such an experiment might keep the severity of character affect the same
and vary the strength of the moral transgression.
112
Figure 5.1: Evaluation places forward constraints on simulation.
5.5 Character manipulation
In each of the three studies reported in chapters 2‐4, characters were manipulated
in terms of morality, that is to say they were either ‘good’ or ‘bad’. The recorded
corrugator EMG showed that bad characters consistently evoked higher corrugator
activity than good characters did across all three studies. This result was clearest in
studies 1 and 2 (see figure 2.1 and 3.1). In these first two studies the character’s
moral status was manipulated with a narrative description of each character either
behaving morally or immorally in a concrete situation. The morality judgement was
implicit insofar as the character was never labelled as good or bad. In contrast,
study 3 manipulated character morality with an explicit statement that they were
‘good’ or ‘bad’, without any narrative framing. This propositional way of
113
establishing character morality evoked a markedly smaller increase in corrugator
activity that was only marginally significantly higher for immoral characters (see
figure 4.3).
The marginal difference in effectiveness of the character manipulations
might be due to the narrative vs. propositional nature of the manipulation.
Showing the moral status by describing concrete situations and behaviour is
arguably a richer emotionally competent stimulus compared to telling people a
character is ‘good’ or ‘bad’. The degree of explicitness likely also plays a role.
Regardless, the aim of this manipulation was to influence the way subsequent
affective linguistic stimuli were evaluated. Given the way patterns of corrugator
activity during descriptions of affective states and events befalling those same
characters were shown to depend heavily on character morality, the morality
manipulation was successful.
In study 3, the morality manipulation was contrasted with a character
manipulation based on the minimal group paradigm. This minimal group
manipulation was hypothesised to evoke weaker group‐dependent evaluation. The
corrugator response confirmed a smaller effect of minimal group status compared
moral group status insofar as it did not evoke any differential corrugator activity. In
a future experiment, it would be interesting to investigate the effects of minimal‐
group based character manipulations using full narrative description of concrete
behaviour as in study 1 and 2. It may well be the case that in such an experiment,
the minimal group distinction will elicit stronger group‐dependent evaluation than
we saw in study 3, but we would still expect this group‐dependent evaluation to be
less strong that for the morality‐based distinction.
5.6 Concluding Remarks
In this dissertation I have attempted to shed light on previous evidence suggesting
that facial EMG activity during language comprehension reflects simulation in
114
service of affective language comprehension. The three studies presented here
provide clear evidence against an account of corrugator activity in terms of
simulation alone. Simulation alone can not account for the patterns that emerge in
response targeted affective words and subclauses in naturalistic stimuli that
engage the reader with the material through evaluation. Having said that, as
discussed in section 5.4 above, a reconceptualisation of simulation processes as
more dynamic and subject to context‐based forward constraints might also account
for the patterns found. Such an account can only be accepted if the complex
evaluation explanation can be satisfactorily dicarded. Regardless of the final word
on simulation, future research on affective language comprehension within a
grounded cognition framework would do well to consider the complexity of
affective meaning in more ecologically valid language stimuli.
115
References
Alves, H., Koch, A., & Unkelbach, C. (2017). Why good is more alike than
bad: Processing implications. Trends in Cognitive Science.
Anderson, M. L. (2010). Neural reuse: A fundamental organizational
principle of the brain. Behavioral and Brain Sciences, 33, 245‐266.
Aziz‐Zadeh, L., Fiebach, C. J., Naranayan, S., Feldman, J., Dodge, E., & Ivry, R.
B. (2008). Modulation of the FFA and PPA by language related to
faces and places. Social Neuroscience, 229‐238.
Aziz‐Zadeh, L., Wilson, S. M., Rizzolatti, G., & Iacobini, M. (2006). Congruent
embodied representations for visually presented actions and
linguistic phrases describing actions. Current Biology, 16, 1818‐
1823.
Bal, M. (2009). Narratology: Introduction to the theory of narrative (3rd
ed.). Toronto: University of Toronto Press Incorporated.
Barsalou, L. W. (1999). Perceptual Symbol Systems. Behavioural and Brain
Sciences, 577‐660.
Barsalou, L. W. (2003b). Situated simulation in the human conceptual
system. Language and Cognitive Processes, 18, 513‐562.
Barsalou, L. W. (2008). Grounded Cognition. Annual Review of Psychology,
617‐645.
Barsalou, L. W. (2016). On staying grounded and avoiding quixotic dead
ends. Psychonomic Bulletin & Review, 23, 1122‐142.
Borghi, A. M., & Riggio, L. (2009). Sentence comprehension and simulation
of objects' temporary, canonical and stable affordances. Brain
Research, 117‐128.
Borreggine, K. L., & Kaschak, M. P. (2006). The action‐sentence
compatibility effect: It's all in the timing. Cognitive Science, 1097‐
1112.
116
Cacioppo, J. T., Berntson, G. G., Norris, C. J., & Higgins, T. E. (2012). The
Evaluative Space Model. In P. M. Van Lange, A. W. Kruglanski, & T.
E. Higgins (Eds.), Handbook of Theories of Social Psychology (Vol. 1,
pp. 42‐64). SAGE Publications Ltd.
doi:http://dx.doi.org/10.4135/9781446249215
Cacioppo, J. T., Petty, R. E., & Morris, K. J. (1985). Semantic, evaluative, and
self‐referent processing: memory, cognitive effort, and
somatovisceral activity. Psychphysiology, 22(4), 371‐384.
Chen, M., & Bargh, J. A. (1999). Consequences of automatic evaluation:
Immediate behavioural predisposition to approach ro avoid the
stimulus. Personality and Social Psychology Bulletin, 215‐224.
Chory, R. M. (2013). Differences in television viewers' involvement :
Identification with and attrraction to liked, disliked, and neutral
characters. Communication Research Reports, 30(4), 293‐305.
doi:http://dx.doi.org/10.1080/08824096.2013.837041
Chow, H.‐M., Mar, R. A., Xu, Y., Liu, S., Wagage, S., & Braun, A. R. (2014).
Embodied comprehension of stories: Interaction between language
systems and modality‐specific neural systems. Journal of Cognitive
Neuroscience, 279‐295.
Cikara, M., & Fiske, S. T. (2012). Stereotypes and Schadenfreude: Affective
and physiological markers of pleasure at outgroup misfortunes.
Social Psychological and Personality Science, 3(1), 1‐16.
Cikara, M., Bruneau, E. G., & Saxe, R. R. (2012). Us and them: intergroup
failures of empathy. Current Directions in Psychological Science,
20(3), 149‐153.
Cikara, M., Bruneau, E., Van Bavel, J. J., & Saxe, R. (2014). Their pain gives us
pleasure: How intergroup dynamics shape empathic failures and
counter‐empathic responses. Journal of Experimental Psychology,
55, 110‐125.
117
Cutting, J., & Iricinschi, C. (2015). Re‐presentations of space in Hollywood
movies: An event‐indexing analysis. Cognitive Science, 39(2), 434‐
456.
Dal Cin, S., Zanna, M. P., & Fong, G. T. (2004). Narrative persuasion and
overcoming resistance. In E. S. Knowles, & J. A. Linn, Resistance &
Persuasion (pp. 175‐193). London: Lawrence Erlbaum Associates
Publishers.
Darwin, C. (1872). The expression of emotion in man and animals. London:
John Murray.
Darwin, C., Ekman, P., & Prodger, P. (1998). The expression of the emotions
in man and animals (3rd ed.). London: Harper Collins.
Davis, M. H. (1983). Measuring individual differences in empathy: Evidence
for a multidmensional approach. Journal of Personality and Social
Psychology, 44, 113‐126.
de Koning, B. B., Wassenburg, S. I., Bos, L. T., & van der Schoot, M. (2016).
Size does matter: Implied object size is mentally simulated during
language comprehension. Discourse Processes, 1‐11.
de Vega, M., Robertson, D. A., Glenberg, A. M., Kaschak, M. P., & Rinck, M.
(2004). On doing two things at once: Temporal constraints on
actions in language comprehension. Memory & Cognition, 1033‐
1043.
Decety, J., & Cowell, J. (2014). The complex relation between morality and
empathy. Trends in Cognitive Science, 18(7), 337‐339.
doi:10.1016/j.tics.2014.04.008
Diehl, M. (1990). The minimal group paradigm: Theoretical explanations
and empirical findings. European Review of Social Psychology, 1(1),
263‐292.
118
Dimberg, U., Thunberg, M., & Elmehed, K. (2000). Unconscious facial
reactions to emotional facial expressions. Psychological Science, 86‐
89.
Ditman, T., Brunyé, T. T., Mahoney, C. R., & Taylor, H. A. (2010). Simulating
an enactment effect: Pronouns guide action simulation during
narrative comprehension. Cognition, 172‐178.
Dunbar, R. I. (2004). Gossip in evolutionary perspective. Review of General
Psychology, 8(2), 100‐110. doi:10.1037/1089‐2680.8.2.100
Ekman, P. (1999). Basic Emotions. In T. Dalgleish, & M. J. Power (Eds.),
Handbook of Cognition and Emotion (pp. 45‐60). New York: John
Wiley & Sons Ltd.
Ekman, P., Friesen, W. V., & Ellsworth, P. (1982). What emotion categories
or dimensions can observers judge from facial behavior? In P.
Ekman (Ed.), Emotion in the human face (pp. 39‐55). New York:
Cambridge University Press.
Ekman, P., Friesen, W. V., & Ellsworth, P. (1982). What emotions categories
or dimensions can be observers judge from facial behaviour? In P.
Ekman, Emotion in the human face (pp. 39‐55). New York:
Cambridge University Press.
Enfield, N. (2013). Relationship Thinking. New York City: Oxford University
Press.
Feather, N. T., & Nairn, K. (2005). Resentment, envy, schadenfreude, and
sympathy: Effects of own and other's deserved or undeserved
status. Australian journal of Psychology, 57(2), 87‐102.
Fincher‐Kiefer, R., & D'Agostino, P. R. (2004). The role of visuospatial
resources in generating predictive and bridging inferences.
Discourse Processes, 205‐224.
119
Fino, E., Menegatti, M., Avenanti, A., & Rubini, M. (2016). Enjoying vs.
Smiling: Facial muscular activation in response to emotional
language. Biological Psychology, 118, 126‐135.
Fludernik, M. (1996). Towards a 'natural' narratology. London: Routledge.
Fludernik, M. (2003). Natural narratology and cognitive paramters. In D.
Herman, Narrative Theory and the Cognitive Sciences (pp. 243‐267).
Stanford: CSLI Publications.
Foroni, F., & Semin, G. R. (2009). Language that puts you in touch with your
bodily feelings: The multimodal responsiveness of affective
expressions. Psychological Science, 974‐980.
Foroni, F., & Semin, G. R. (2013). Comprehension of action negation inolves
inhibitory simulation. Frontiers in Human Neuroscience, 1‐7.
doi:10.3389/fnhum.2013.00209
Gibbs, R. W. (2006). Embodiment and cognitive science. New York:
Cambridge University Press.
Gibbs, R. W. (2012). Walking the walk while talking the talk: Embodied
interpretation of metaphorical narratives. Journal of
Psycholinguistic Research, 363‐378.
Glenberg, A. M. (1997). What memory is for. Behavioral and Brain Sciences,
1‐55.
Glenberg, A. M., & Kaschak, M. P. (2002). Grounding language in action.
Psychonomic Bulletin & Review, 558‐565.
Glenberg, A. M., Sato, M., Cattaneo, L., Riggio, L., Palumbo, D., & Buccino,
G. (2008). Procesisng asbtract language modulates motor system
activity. Quarterly JOurnal of Experimental Psychology, 901‐919.
Glenberg, A. M., Webster, B. J., Mouilso, E., Havas, D., & Lindeman, L. M.
(2009). Gender, emotion, and the embodiment of language
comprehension. Emotion Review, 151‐161.
120
González, J., Barros‐Loscertales, A., Pulvermüller, F., Meseguer, V., Sanjuán,
A., Belloch, V., & Ávila, C. (2006). Reading cinnamon activates
olfactory brain regions. NeuroImage, 32, 906‐912.
Gothard, K. M. (2014). The amygdalo‐motor pathways and the control of
facial expressions. Frontier in Neuroscience, 1‐7.
Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing inferences
during narrative text comprehension. Psychological Review, 33‐50.
Graham, J., Nosek, B. A., Haidt, J., Iyer, R., Koleva, S., & Ditto, P. H. (2011).
Mapping the moral domain. Journal of Personality and Social
Psychology, 101(2), 366‐385.
Green, M. C., & Carpenter, J. M. (2011). Transporting into narrative worlds:
New directions for the scientific study of literature. Scientific Study
of Literature, 113‐122.
Green, M. C., Brock, T. C., & Kaufman, G. F. (2004). Understanding media
enjoyment: the role fo transportation into narrative worlds.
Communication Theory, 311‐327.
Greene, J. D. (2014). The cognitive neuroscience of moral judgment and
decision‐making. In M. S. Gazzaniga (Ed.), The Cognitive
Neuroscience V (pp. 1013‐1024). Cambridge, Massachusetts, USA:
MIT Press.
Haidt, J. (2012). The Righteous Mind: Why good people are divided by
politicis and religion. London, United Kingdom: Penguin Books Ltd.
Harnad, S. (1990). The symbol grounding problem. Physica D: Nonlinear
Phenomena, 42(1‐3), 335‐346.
Hauk, O., Johnsrude, I., & Pulvermuller, F. (2004). Somatotopic
representation of action words in human motor and premotor
cortex. Neuron, 301‐307.
121
Havas, D. A., & Matheson, J. (2013). The functional role of the periphery in
emotional language processing. Frontiers in Psychology, 1‐16.
Havas, D. A., Glenberg, A. M., & Rinck, M. (2007). Emotion simulation
during language comprehension. Psychonomic Bulletin & Review,
14(3), 436‐441.
Havas, D. A., Glenberg, A. M., Gutowski, K. A., Lucarelli, M. J., & Davidson, R.
J. (2010). Cosmetic use of botulinum toxin‐a affects processing of
emotional language. Psychological Science, 895‐900.
Heider, F., & Simmel, M. (1944). An experimental study of apparent
behavior. The American Journal of Psychology, 57(2), 243‐259.
Hennenlotter, A., Dresel, C., Castrop, F., Ceballos, A. O., Wohlschläger, A.
M., & Haslinger, B. (2009). The link between facial feedback and
neural activity within central circuitries of emotion: New insights of
botulinum toxin‐induced denervation of frown muscles. Cerebral
Cortex, 19, 537‐542.
Herman, D. (2009). Basic elements of narrative. Chichester: Wiley‐Blackwell.
Hoeken, H., & Sinkeldam, J. (2014). The role of identification and perception
of just outcome in evoking emotions in narrative persuasion.
Journal of Communication, 64(5), 935‐955. doi:10.1111/jcom.12114
Hofman, D., Bos, P. A., Schutter, D. J., & van Honk, J. (2012). Fairness
modulates non conscious facial mimicry in women. Proceedings of
the Royal Society B, 3535‐3539.
Horchak, O. V., & Giger, J.‐C. (2014). Simulation of metaphorical actions and
discourse comprehension. Metaphor and Symbol, 1‐22.
Horchak, O. V., Giger, J.‐C., & Pochwatko, G. (2014). Discourse
comprehension and simulation of positive emotions. Psicológica,
17‐37.
122
Horchak, O. V., Giger, J.‐C., Cabral, M., & Pochwatko, G. (in press). From
demonstration to theory in embodied language comprehension: A
review. Cognitive Systems Research, xxx‐xxx.
Isenberg, N., Silbersweg, D., Engelien, A., Emmerich, S., Malavade, K.,
Beattie, B., & Leon, A. C. (1999). Linguistic threat activates the
human amygdala. Proceedings of the National Academy of Sciences,
10456‐10459.
Kaschak, M. P., Madden, C. J., Therriault, D. J., Yaxley, R. H., Aveyard, M., &
Blanchard, A. A. (2005). Perceptions of motion affects language
processing. Cognition, 79‐89.
Kaschak, M. P., Zwaan, R. A., Aveyard, M., & Yaxley, R. H. (2006). Perception
of auditory motion affects language processing. Cognitive Science,
733‐744.
Keltner, D., & Ekman, P. (2000). Facial expression of emotion. In M. Lewis, &
J. Haviland‐Jones (Eds.), Handbook of Emotions (pp. 236‐249). New
York: Guildford Publications, Inc.
Kiefer, M., & Pulvermüller, F. (2012). Conceptual representations in mind
and brain: Theoretical developments, current evidence and future
directions. Cortex, 805‐825.
Kiefer, M., Sim, E.‐J., Herrnberger, B., Grothe, J., & Hoenig, K. (2008). The
sound of concepts:four markers for a link between auditory and
conceptual brain systems. The Journal of Neuroscience, 28(47),
12224‐12230.
Kintsch, W., & van Dijk, T. A. (1978). Toward a model of comprehension and
production. Psychological Review, 363‐394.
Künecke, J., Sommer, W., Schacht, A., & Palazova, M. (2015). Embodied
simulation of emotional valence: Facial muscle responses to
abstract and concrete words. Psychophysiology, 1590‐1598.
123
Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: Chicago
University Press.
Landau, A. N., Aziz‐Zadeh, L., & Ivry, R. B. (2010). The influence of language
processing on perception: listening to sentences about faces affects
the perception of faces. Journal of Neuroscience, 15254‐15261.
Larsen, J. T., Norris, C. J., & Cacioppo, J. T. (2003, September). Effects of
positive and negative affect on electromyographic activity over
zygomaticus major and corrugator supercilii. Psychophysiology,
776‐785.
Leach, C. W., & Spears, R. (2009). Dejection at in‐group defeat and
schadenfreude toward second‐ and third‐party out‐groups.
Emotion, 9(5), 659‐665.
LeDoux, J. (1996). The emotional brain. New York: Simon & Schuster.
Leshinskaya, A., & Caramazza, A. (2016). For cognitive neuroscience of
concepts: Moving beyod the grounding issues. Psychonomic Bulletin
& Review, 23(4), 991‐1001.
Leuthold, H., Kunkel, A., Mackenzie, I. G., & Filik, R. (2015). Online
processing of moral transgressions: ERP evidence for spontaneous
evaluation. Social and Cognitive Affective Neuroscience, 10(8),
1021‐1029.
Luo, S., Li, B., Ma, Y., Zhang, W., Yi, R., & Han, S. (2015). Oxytocin receptor
gene and racial ingroup bias in empathy‐related brain activity.
NeuroImage, 110, 22‐31.
Mahon, B. Z., & Caramazza, A. (2008). A critical look at the embodied
cognition hypthesis and a new proposal for grounding conceptual
content. Journal of Physiology‐Paris, 59‐70.
Mahon, B. Z., & Hickok, G. (2016). Arguments about the nature of concepts:
Symbols, embodiment, and beyond. Psychonomic Bulletin &
Review, 23(4), 941‐958.
124
Masson, M. E., Bub, D. N., & Warren, C. M. (2008b). Kicking calculators:
Contribution of embodied representations to sentence
comprehension. Journal of Memory and Language, 256‐265.
Mirman, D. (2015, January 6). Growth curve analysis. Retrieved from
Language & Cognitive Dynamics Laboratory: www.danmirman.org
Montalan, B., Lelard, T., Godefroy, O., & Mouras, H. (2012). Behavioral
investigation of the influence of social categorization on empathy
for pain: a minimal group paradigm study. Frontiers in Psychology:
Cognitive Science, 1‐5.
Montoro, P. R., Contreras, M. J., & Elosúa, M. R. (2015). Cross‐modal
metaphorical mapping of spoken emotion words onto vertical
space. Frontiers in Psychology, 6, 1‐13.
Morrison, S., Decety, J., & Molenberghs, P. (2012). The neuroscience of
group membership. Neuropsychologica, 50, 2114‐2120.
Moscoso del Prado Martin, F., Hauk, O., & Pulvermuller, F. (2006). Category
specificity in the processing of color‐related and form‐related
words: an ERP study. Neuroimage, 29‐37.
Moseley, R., Carota, F., Hauk, O., Mohr, B., & Pulvermüller, F. (2012). A role
for the motor system in binding abstract emotional meaning.
Cerebral Cortex, 1634‐1647.
Moyer‐Gusé, E. (2008). Toward a theory of entertainment persuasion:
Explaining the persuasive effects of entertainment‐education
messages. Communication Theory, 407‐425.
Moyer‐Gusé, E., & Nabi, R. L. (2010). Explaining the effects of narrative in an
entertainment television program: Overcoming resistance to
persuasion. Human Communication Research, 26‐52.
Myachykov, A., Ellis, R., Cangelosi, A., & Fischer, M. (2013). Visual and
linguistic cues to graspable objects. Experimental brain Research,
549‐559.
125
Niedenthal, P. M., Winkielman, P., Mondillon, L., & Vermeulen, N. (2009).
Embodiment of emotion concepts. Journal of Personality and Social
Psychology, 1120‐1136.
Oatley, K. (1995). A taxonomy of the emotions of literary response and a
theory of identification in fictional narrative. Poetics, 23(1), 53‐74.
doi:http://dx.doi.org/10.1016/0304‐422X(94)P4296‐S
Oatley, K. (1999). Meeting of minds: Dialogue, sympathy, and
identificationin reading fiction. Poetics, 28(5‐6), 439‐454.
Oberman, L. M., Winkielman, P., & Ramachandran, V. S. (2007). Face to
face: Blocking expression‐specific muscles can selectively impair
recognition of emotional faces. Social Neuroscience, 167‐178.
Ortony, A., & Turner, T. J. (1990). What's basic about basic emotions.
Psychological Review, 97(3), 315‐331.
Pecher, D., Zeelenberg, R., & Barsalou, L. W. (2003). Verifying different
modality properties for concepts produces switching costs.
Psychological Science, 119‐124.
Peck, R., & Devore, J. (2008). Statistics: The Exploration and Analysis of Data
(6th ed.). London, United Kingdom: Thomson Higher Learning.
Porter Abbott, H. (2008). The cambridge introduction to narrative. New
York: Cambridge University Press.
Posner, J., Russell, J. A., & Peterson, B. S. (2005). The circumplex model of
affect: An integrative approach to affective neuroscience, cognitive,
development, and psychopathology. Development &
Psychopathology, 17(3), 715‐734.
Pulvermüller, F. (2005). Brain mechanisms linking language and action.
Nature Reviews Neuroscience, 6, 576‐582.
Rapp, D. N., & Taylor, H. A. (2004). Interactive dimensions in the
construction of mental representations for text. Journal of
126
Experimental Psychology; Learning, memory, and Cognition, 988‐
1001.
Ratner, K. G., & Amodio, D. M. (2012). Seeing "us vs. them": Minimal group
effects on the neural encoding of faces. Journal of Experimental
Social Psychology, 49, 298‐301.
Rubin, D. C. (2006). The basic‐systems model of episodic memory.
Perspectives on Psychological Science, 277‐311.
Rueschemeyer, S. A., van Rooij, D., Lindemann, O., Willems, R. M., &
Bekkering, H. (2010). The function of words: Distinct neural
correlates for words denoting differently manipulable objects.
Journal of Cognitive Neuroscience, 1844‐1851.
Ryan, M. L. (2007). Toward a definition of narrative. In D. Herman, The
Cambridge Companion to Narrative (pp. 22‐35). Cambridge:
Cambridge University Press.
Samur, D., Lai, V. T., Hagoort, P., & Willems, R. M. (2015). Emotional context
modulates embodied metaphor comprehension. Neuropsychologia,
108‐114. doi:10.1016/j.neuropsychologia.2015.10.003
Sanford, A. J., & Emmott, C. (2012). Mind, brain, and narrative. New York:
Cambridge University Press.
Santana, E., & de Vega, M. (2011). Metaphors are embodied, and so are
their literal counterparts. Frontiers in Cognition, 2‐90.
Schenk, C. H. (2015). REM sleep behaviour disorder. In S. Chokroverty, & M.
Billiard (Eds.), Sleep Medicine (pp. 391‐405). New York: Springer
Science + Business Media LLC.
Schuil, K. D., Smits, M., & Zwaan, R. A. (2013). Sentential context modulates
the involvement of the motor cortex in action language processing:
an fMRI study. Frontiers in Human Neuroscience, 1‐13.
doi:10.3389/fnhum.2013.00100
127
Sell, A. J., & Kaschak, M. P. (2011). Processing time shift affects the
execution of motor responses. Brain and Language, 39‐44.
Sell, A. J., & Kaschak, M. P. (2012). The comprehension of sentences
involving quantity information affects responses on the up‐down
axis. Psychonomic Bulletin and Review, 708‐714.
Shiota, M. N., & Kalat, J. W. (2012). Emotion. Wadsworth, Cengage
Learning.
Simmons, W. K., Ramjee, V., Beauchamp, M. S., McRae, K., Martin, A., &
Barsalou, L. W. (2007). A common neural substrate for perceiving
and knowing about colour. Neuropsychologia, 2802‐2810.
Singer, T., Seymour, B., O'Doherty, J. P., Stephan, K. E., Dolan, R. J., & Frith,
C. D. (2006). Empathic neural responses are modulated by the
perceived fairness of others. Nature, 439, 466‐469.
Slater, M. D., & Rouner, D. (2002). Enterainment‐education and elaboration
likelihood: Understanding the processing of narrative persuasion.
Communication Theory, 173‐191.
Speer, N. K., Reynolds, J. R., Swallow, K. M., & Zacks, J. M. (2009). Reading
stories activates neural representations of visual and motor
experiences. Psychological Science, 20(8), 989‐999.
Speer, N. K., Zacks, J. M., & Reynolds, J. R. (2007). Human brain activity
time‐locked to narrative event boundaries. Psychological Science,
18(5), 449‐455.
Spivey, M. J., & Geng, J. J. (2001). Oculomotor mechanisms activated by
imagery and memory: Eye movement to absent objects.
Psychological Research, 235‐241.
Strack, F., Martin, L. L., & Stepper, S. (1988). Inhibiting and facilitating
conditions of the human smile: A nonobtrusive test fo the facial
feedback hypothesis. Journal of Personality and Social Psychology,
54, 768‐777.
128
't Hart, B., Struiksma, M. E., van Boxtel, A., & Van Berkum, J. J. A. (in
preparation a). Emotion in stories: A facial EMG study on simulation
vs. moral evaluation.
't Hart, B., Struiksma, M. E., van Boxtel, A., & Van Berkum, J. J. A. (in
preparation b). Simulation and evaluation of emotions in stories:
Online processing of character affect.
Tajfel, H., Billig, M., Bundy, R., & Flament, C. (1971). Social categorization
and intergroup behaviour. European Journal of Social Psychology, 1,
149‐178.
Tappin, B. M., & McKay, R. T. (2016). The illusion of moral superiority. Social
Psychological and Personality Science, 1‐9.
Tassinary, L. G., & Cacioppo, J. T. (1992). Unobservable facial actions and
emotion. Psychological Science, 28‐33.
Tassinary, L. G., Cacioppo, J. T., & Vanman, E. J. (2000). The skeletomotor
system: Surface electromyography. In J. T. Cacioppo, L. G. Tassinary,
& G. G. Berntson, Handbook of Psychophysiology (pp. 267‐299).
Cambridge: Cambridge University Press.
Taylor, L. J., & Zwaan, R. A. (2008). Motor resonance and linguistic focus.
Quarterly Journal of Expperimental Psychology Section A‐HUman
Experimental P, 896‐904.
Tettamanti, M., Buccino, G., Saccuman, M. C., Gallese, V., Danna, M., Scifo,
P., . . . Perani, D. (2005). Listening to action‐related sentences
activates fronto‐parietal motor circuits. Journal of Cognitive
Neuroscience, 273‐281.
Tian, Q., & Hoffner, C. A. (2010). Parasocial interaction with liked, disliked,
and neutral characters on a popular TV series. Mass
Communication and Society, 13(3), 250‐269.
doi:http://dx.doi.org/10.1080/15205430903296051
129
Topolinski, S., Likowski, K. U., Weyers, P., & Strack, F. (2009). The face of
fluency: Semantic coherence automatically elicits a specific pattern
of facial muscle reactions. Cognition & Emotion, 23(2), 260‐271.
Topolinsky, S., & Strack, F. (2015). Corrugator activity confirms immediate
negative affect in surprise. Frontiers in Psychology, 6, 1‐8.
Turing, A. M. (1950). Computing machinery and intelligence. Mind, 433‐460.
Turner, T. J., & Ortony, A. (1992). Basic emotions: Can conflicting criteria
converge. Psychological Review, 99(3), 566‐571.
Van Berkum, J. J. A. (2008). Understanding sentences in context: What brain
waves can tell us. Current Directions in Psychological Science, 17(6),
376‐380. doi:10.1111/j.1467‐8721.2008.00609.x
Van Berkum, J. J. A. (2016a). Language comprehension and emotion: Where
are the interfaces, and who cares? In G. de Zubicaray, & N. O.
Schiller (Eds.), Oxford Handbook of Neurolinguistics. Oxford
University Press.
Van Berkum, J. J. A.(in press). Language comprehension, emotion and
sociality: Aren't we missing something? In Rueschemeyer, & Gaskell
(Eds.), Oxford Handbook of Psycholinguistics. Oxford: OUP.
Van Berkum, J. J. A., Holleman, B., Nieuwland, M., Otten, M., & Murre, J.
(2009). Right or wrong? The brain's fast response to moralliy
objectionable statements. Psychological Science, 20(9), 1092‐1099.
doi:10.1111/j.1467‐9280.2009.02411.x
van Boxtel, A. (2010). Facial EMG as a tool for inferring affective states.
Proceedings of Measuring Behaviour 2010 (pp. 104‐108).
Eindhoven: Measuring Behaviour 2010.
Vermeulen, N., Niedenthal, P. M., & Luminet, O. (2007). Switching between
sensory and affective systems incurs processing costs. Cognitive
Science, 183‐192.
130
Vigliocco, G., Meteyard, L., Andrews, M., & Kousta, S. (2009). Toward a
theory of semantic representation. Language and Cognition, 219‐
247.
Vossen, H. G., Piotrowski, J. T., & Valkenburg, P. M. (2015). Development of
the Adolescent Measure of Empathy and Sympathy (AMES).
Personality and Individual Differences, 66‐71.
Wagenmakers, E.‐J., Beek, T., Dijkhoff, L., Gronau, Q. F., Acosta, A., Adams,
R. B., . . . Chasten, K. T. (2016). Registered Replication Report:
Strack, Martin, & Stepper (1988). Perspective on Psychological
Science, 11(6), 917‐928.
Whitney, C., Huber, W., Klann, J., Weiss, S., Krach, S., & Kircher, T. (2009).
Neural correlates of narrative shifts during auditory story
comprehension. NeuroImage, 47(1), 360‐366.
Wicker, B., Keysers, C., Plailly, J., Royet, J. P., Gallese, V., & Rizzolatti, G.
(2003). Both of us disgusted in my insula: The common neural basis
of seeing and feeling disgust. Neuron, 655‐664.
Willems, R. M., & Cassanto, D. (2011). Flexibility in embodied language
understanding. Frontiers in Psychology, 1‐11.
Willems, R. M., Hagoort, P., & Cassanto, D. (2010b). Body‐specific
representation of action verbs: Neural evidence form right‐ and
left‐handers. Psychological Science, 67‐74.
Wundt, W. (1977 (1907)). Lectures on human and animal psychology.
Washington, D.C.: University Publications of America.
Yee, E., & Thompson‐Schill, S. L. (2016). Putting concepts into context.
Psychonomic Bulletin & Review, 23, 1015‐1027.
Zacks, J. M., Speer, N. K., Swallow, K. M., & Maley, C. J. (2010). The brain’s
cutting‐room floor: Segmentation of narrative cinema. Frontiers in
HUman Neuroscience, 4, 1‐15.
131
Zwaan, R. A. (2004). The Immersed Experiencer: Toward an embodied
theory of language comprehension. The Psychology of Learning and
Motivation, 35‐62.
Zwaan, R. A. (2009). Mental simulation in language comprehension and
social cognition. European Journal of Social Psychology, 1142‐1150.
Zwaan, R. A. (2014). Embodiment and language comprehension: Reframing
the discussion. Trends in Cognitive Sciences, 1‐6.
doi:10.1016/j.tics.2014.02.008
Zwaan, R. A. (2016). Situation models, mental simulations, and abstract
concepts in discourse comprehension. Psychonomic Bulletin &
Review, 23(4), 1028‐1034.
Zwaan, R. A., & Kaschak, M. P. (2008). Language in the brain, body, and
world. In P. Robbins, & M. Aydede (Eds.), The Cambridge Handbook
of Situated Cognition (pp. 368‐381). Cambridge University Press.
Zwaan, R. A., & Madden, C. J. (2005). Embodied sentence comprehension.
In D. P. Pecher, & R. A. Zwaan (Eds.), Grounding Cognition: The role
of perception and action in memory, language, and thinking (pp.
224‐245). New York: Cambridge University Press.
Zwaan, R. A., & Pecher, D. (2005). Introduction to grounding cognition. In
Various, R. A. Zwaan, & D. Pecher (Eds.), Grounding Cognition (pp.
1‐7). Cambridge, New York, United States of America: Cambridge
University Press.
Zwaan, R. A., & Pecher, D. (2012). Revisiting mental simulation in language
comprehension: Six replication attempts. PLoS ONE.
Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language
comprehension and memory. Psychological Bulletin, 162‐185.
Zwaan, R. A., & Taylor, L. J. (2006). Seeing, acting, understanding: Motor
resonance in language comprehension. Journal of Experimental
Psychology‐General, 1‐11.
132
Zwaan, R. A., Madden, C. J., Yaxley, R. H., & Aveyard, M. (2004). Moving
words: Dynamic mental representations in language
comprehension. Cognitive Science, 611‐619.
Zwaan, R. A., Stanfield, R. A., & Yaxley, R. H. (2002). Language
comprehenders mentally represent the shapes of objects.
Psychological Science, 168‐171.
Zwaan, R. A., Taylor, L. J., & de Boer, M. M. (2010). Motor resonance as a
function of narrative time: Further tests of linguistic‐focus theory.
Brain & Language, 143‐149.
133
Supplementary Information A
A1: Study 1
1
Sandra heeft al een tijdje een conflict met haar buren. Ze zet haar auto vaak op de parkeerplek voor de deur van haar buren. Haar buurman en buurvrouw beweren echter dat die plek speciaal voor hun auto is. Op een dag hoort Sandra van de andere buren dat haar buurman verongelukt is. Sandra aarzelt niet en loopt direct naar het huis van haar buren. Ze belt aan en haar buurvrouw doet de deur open.
Sandra bijt haar toe dat haar man toch een vervelende vent was en het waarschijnlijk verdiende om dood te gaan.
Sandra loopt weer naar huis en gaat verder met koken. Die avond pakt Sandra haar tablet en leest het financiële nieuws. Sandra heeft kort geleden veel van haar spaargeld in aandelen gestoken. Ze houdt sindsdien de markt goed in de gaten en nu maakt ze zich zorgen. Ze heeft slechte berichten gehoord en checkt de koersen van de aandelen in haar portfolio.
… Sandra is dolblij
or
woedend als ze de gegevens doorleest
en haar aandelen in waarde verdubbbeld zijn
or
en haar aandelen niks meer waard blijken te zijn
2
De afdeling waar Wesley werkt heeft het heel erg druk. De opdrachten blijven maar binnenstromen. Daarom is er kort geleden een nieuwe medewerker aangenomen om het team te versterken. De nieuwe medewerker is een jong en knap meisje dat pas afgestudeerd is. Op een dag komt Wesley haar
Wesley gaat achter haar staan, vraagt wat er is, en streelt zogenaamd per ongeluk de billen van het meisje.
Grijnzend werkt Wesley die middag verder en gaat om vijf uur naar huis. Thuis aangekomen parkeert hij zijn auto en loopt naar de voordeur van zijn huis. Wesley hoopt dat bij de post de dvd zit die hij besteld had. Hij had de hele avond vrijgehoud
… Wesley is blij
or
geërgerd wanneer hij de post doorbladert
en de dvd erbij zit en hij dus door kan kijken
or
de dvd er niet bijzit, maar wel een boete
134
alleen tegen in het kopieerhok. Ze staat bij het kopieerapparaat en heeft duidelijk ruzie met het apparaat.
en om het nieuwe seizoen van de tv serie te kijken. Wesley doet de deur open en ziet een stapel post op de deurmat.
3
Jeffrey komt na een dag hard werken thuis. Net op het moment dat hij de voordeur achter zich dicht trekt begint het te onweren. Het bliksemt en begint hard te regenen. Dan ziet Jeffrey dat er die dag post is bezorgd bij hem. Tussen de folders en rekeningen die zijn bezorgd ziet Jeffrey een rouwkaart liggen. De kaart blijkt verkeerd bezorgd te zijn en eigenlijk bestemd voor zijn overburen.
Jeffrey heeft geen zin om zich nat te laten regenen en verscheurt de rouwkaart en gooit hem in de prullenbak.
Het noodweer blijft die dag aanhouden. Jeffrey hoort dat het KNMI code rood afgeeft en aanraadt binnen te blijven. Het begint hard te hagelen en Jeffrey ziet buiten enorme hagelstenen uit de lucht vallen. Een hagelsteen zo groot als een golfbal landt in Jeffreys tuin en Jeffrey vreest voor zijn nieuwe auto. Zodra het droog is gaat Jeffrey de schade opnemen.
… Jeffrey voelt
opluchting
or
frustratie wanneer hij bij de auto aan komt lopen
en deze geen beschadigingen blijkt te hebben
or
en die onder de kleine deuken zit van de hagelstenen
4
Emma loopt over de Dam richting de Kalverstraat. Ze moet een verjaardagscadeautje halen voor een vriend. Het is gelukkig niet heel druk, maar het loopt al tegen sluitingstijd. Op de hoek van de Kalverstraat zit een blinde straatmuzikant. Hij speelt de sterren
Emma heeft alleen een tientje, maar de muzikant is zo goed dat ze het hem gunt, hij kan het vast gebruiken.
Later die week is de verjaardag waarvoor ze op pad was. Er zijn al gauw 25 mensen en het feestje is in volle gang. Emma kent echter niet veel van de mensen die er zijn, en bekenden zijn er nog niet. Ze probeert zich in het gesprek
… Emma voelt
blijdschap
or
woede opkomen wanneer het groepje
haar gelijk welkom doet voelen
or
haar aanstaart en vervolgens negeert
135
van de hemel en zingt met een doorleefde stem. Op de gitaarhoes voor hem ligt al heel wat kleingeld en zelfs een paar briefjes.
van een groepje naast zich te mengen. Emma gaat erbij staan en wanneer een stilte valt doet ze ook een duit in het zakje.
5
Ilse staat te wachten bij de pinautomaat. Er staat één man voor haar en niemand achter haar. Terwijl ze wacht kijkt Ilse wat op haar telefoon en bedenkt hoeveel geld ze op zal nemen. Vanuit haar ooghoek ziet Ilse dat de man voor haar bijna klaar is. Ilse stopt haar telefoon weg en pakt haar portemonnee. Als de man wegloopt, ziet Ilse een briefje van 50 vallen maar de man heeft niks door.
Ilse roept de man gauw na en pakt het briefje van 50 op om terug te geven, de man bedankt Ilse uitvoerig.
Later die avond is Ilse onderweg naar huis. Ze bedenkt zich ineens dat ze niks meer voor het ontbijt in huis heeft. Ze kijkt hoe laat het is en gaat meteen wat sneller lopen. Er is gelukkig een Albert Heijn vlakbij, maar die gaat al bijna dicht. Als Ilse de hoek om loopt ziet ze hoe een medewerker een bord binnenhaalt en ze gaat nog sneller lopen.
… Ilse is dankbaar
or
pissig wanneer de medewerker haar ziet
en toch nog even snel binnenlaat
or
haar vuil aankijkt en de deur snel sluit
6
Martijn fietst door de stad naar de sportschool. Hij heeft een lange dag achter de rug en heeft zin om intensief te sporten. Martijn is halverwege wanneer hij door een erg smal steegje fietst. Een bejaarde man loopt langzaam in dezelfde richting door het steegje. In het voorbijgaan schampt Martijn met zijn fiets de
Martijn beseft dat hij fout zat, maar doet alsof er niks gebeurd is en fietst door, laat een ander die ouwe helpen.
De volgende dag staat Martijn vroeg op. Hij heeft een sollicitatiegesprek bij een bedrijf voor een leidinggevende functie. Het bedrijf is gevestigd in een nabijgelegen stad, 20 kilometer verderop. Omdat Martijn geen rijbewijs heeft reist hij met openbaar
… Martijn is euforisch
or
kwaad wanneer hij op het perron komt en
op tijd blijkt te zijn en de juiste trein haalt
or
hoort omroepen dat er geen treinen rijden
136
oude man. De man komt ten val en Martijn hoort hem kermen van de pijn.
vervoer. Hij wil voor de zekerheid een trein eerder nemen en zorgt dat hij op tijd op het station is.
7
Maria loopt met een ingepakt cadeau over straat. Ze is onderweg naar een kraamvisite bij een collega. Maria heeft niks met baby's en gelukkig heeft iemand anders het cadeau uitgezocht. Maria belt aan en het duurt even voordat de deur opengaat. De kamer zit vol met familie en vrienden die ze niet kent. De baby wordt uiteraard meteen getoond en Maria vindt het maar een lelijke baby.
Maria zegt desgevraagd dat het een mormel is en dat ze sowieso niet snapt wat er mooi is aan baby's.
Wanneer Maria bij de kraamvisite vandaan komt loopt ze het centrum in. Ze heeft honger en gaat op zoek naar iets warms voor de lunch. Ze schiet een döner tent binnen en bestelt een broodje döner. Ze gaat zitten en valt meteen aan wanneer het broodje klaar is. Bij de eerste hap valt er gelijk een enorme klodder saus uit de zijkant van het broodje.
… Maria is nijdig
or
opgelucht als de klodder gemorste saus
op haar nieuwe blouse valt en dus verpest is
or
op tafel valt en niet op haar nieuwe blouse
8
Stefan zit achter de computer op het werk. Hij is bezig concertkaartjes te bestellen voor hem en drie collega's in zijn team. Ze gaan wel vaker met z'n allen iets doen buiten het werk om. Het is ondertussen meer een vriendengroep dan collega's. Er is een nieuw teamlid dat er nog niet echt bij hoort. Dat
Stefan nodigt hem uit om mee te gaan, zodat hij zich meer thuis gaat voelen als nieuweling binnen een hecht team.
Tegen half vijf besluit Stefan dat het leuk is geweest voor die week. Hij sluit zijn computer af en pakt zijn spullen bij elkaar. Voor vijven is hij op het station en wacht hij op de trein. Het nadeel van op tijd weggaan op vrijdag is dat hij niet de enige is. Het is hartstikke
… Stefan is boos
or
opgetogen
wanneer in het gedrang voor de inagng
iemand hem zonder pardon omver duwt
or
iemand hem vriendelijk voor laat gaan
137
nieuwe teamlid komt even later binnen en zegt ook fan te zijn van die band.
druk op het perron en hij ziet dat de binnenkomende trein ook al goed vol zit.
9
Sanne zit naar een voorstelling onder regie van een vriend te kijken. Het stuk is ongeveer halverwege en Sanne is al volledig uitgecheckt. De acteurs zijn niet bijster goed en het decor is amateuristisch. Ook het verhaal zelf kan niet echt boeien. Sanne heeft sterk de neiging om halverwege naar huis te gaan. Maar ze blijft toch tot de receptie en komt daar haar vriend, de regisseur, tegen.
Sanne schudt hem de hand en steekt vervolgens van wal en vertelt keihard hoe slecht ze het stuk vond.
Na de receptie loopt Sanne richting het dichtstbijzijnde grote plein. De trams rijden al niet meer en het begint steeds harder te regenen. Eenmaal bij het plein aangekomen wil ze een taxi aanhouden. Er zijn echter weinig taxi's en veel mensen die een taxi willen vanwege de regen. Eindelijk weet ze een taxi aan te houden en deze stopt vlakbij haar.
… Sanne is verontwaardigd
or
opgetogen
wanneer de taxi vlak voor haar stopt
en iemand haar opzij duwt en gauw instapt
or
en ze in kan stappen voor ze doorweekt is
10
Lieke zit met een vriendin op het terras. Het is donderdagmiddag en ze hebben allebei lekker vakantie. Haar vriendin moet naar het toilet, waar het waarschijnlijk stervensdruk zal zijn. Terwijl Lieke alleen zit komt een vriend van haar langsfietsen. Hij vertelt Lieke over
Lieke vertelt eerlijk over het feestje en vraagt of de vriendin het leuk zou vinden om daarheen te gaan.
Later die middag springt Lieke gauw onder de douche thuis. Onder het afdrogen bedenkt ze zich dat ze cash geld nodig zal hebben. Dat herinnert haar eraan dat haar salaris één dezer dagen binnen moet komen. Nadat ze zich heeft aangekleed rommelt ze
… Lieke constateert
nors
or
opgetogen
dat de salarisadminstratie
weer eens heeft gefaald, nog geen geld
or
heeft gezorgd dat haar salaris op tijd is dit keer
138
een feestje vanavond en moet gelijk verder. Lieke wil wel heen, maar ze heeft eigenlijk al plannen gemaakt met die andere vriendin.
in de la om de random reader te vinden. Eenmaal gevonden logt ze gauw in om haar saldo te checken.
11
Sarah heeft de nachtdienst op de verpleegafdeling. Het is tijd om een ronde te lopen en haar collega en zij doen elk de helft. Sarah controleert alle infusen en kijkt of iedereen goed slaapt. Ze gaat stilletjes van kamer tot kamer. Op één van de laatste kamers ziet ze een portemonnee liggen. Hij is van de oudere man die vanmiddag zijn kleinkinderen met hun rapporten op bezoek had.
Sarah kijkt snel om zich heen, haalt er 50 euro uit en vervolgt haar ronde, de man heeft toch niet lang meer.
Aan het einde van haar dienst gaat Sarah gauw richting huis. Eenmaal thuis zet ze een kop thee en zet haar laptop aan. Met de laptop in één hand en de thee in de andere loopt ze richting de woonkamer. Ze ziet door het voorraam dat de zon ondertussen op is gekomen. Dan struikelt ze ineens over de drempel en laat de laptop vallen en morst thee.
… Sarah constateert
pissig
or
opgelucht dat de laptop als ze hem weer aanzet
helemaal niks meer doet, compleet kapot
or
het weer gewoon blijkt te doen, gelukkig!
12
Danny is aan het werk in de werkplaats van zijn fietsenzaak. Hij is bezig een achterwiel te vervangen. Het is halverwege de ochtend en nog lekker rustig. Wanneer de bel gaat maakt Danny snel af wat hij aan het doen was. Hij veegt zijn handen schoon en loopt dan snel naar
Danny schudt meewarig het hoofd, negeert de vrouw straal en gaat weer verder in de werkplaats, domme allochtonen.
Later die dag, na de lunch, worden er onderdelen geleverd. Danny tekent voor ontvangst en begint de pakketjes te sorteren. Er is één bepaald onderdeel waar hij al een hele tijd op zit te wachten. Het is een heel speciaal onderdeel voor een luxe
… Danny voelt
frustratie
or
opluchting dat het binnengekomen onderdeel
weer het verkeerde blijkt, dit kost hem geld
or
nu wel het goede blijkt te zijn, eindelijk!
139
voren. Er staat een vrouw met een hoofddoek te wachten en het blijkt al gauw dat haar Nederlands niet heel goed is.
mountainbike. Het onderdeel is al twee keer verkeerd geleverd en de klant begint ongeduldig te worden.
13
Kirsten hangt wat rond met haar vrienden in het park. Ze drinken biertjes en zijn wat aangeschoten en baldadig. Een van Kirstens vrienden die meedrinkt begint opmerkingen te maken over mensen die langslopen. Eerst zachtjes, maar later steeds harder. Plotseling loopt een donkere jongen langs het groepje vrienden. Kirstens vriend roept een racistische opmerking naar de jongen en daagt hem uit.
Kirsten lacht lekker mee en doet er een schepje bovenop door de jongen ook uit te schelden en hem te bespugen.
De volgende dag gaat Kirsten naar de elektronicazaak. Ze heeft al een tijd een Ultra HD‐tv op het oog en vandaag wil ze hem kopen. De tv is erg duur en Kirsten heeft er lang voor gespaard. Ze heeft de achterbank van haar auto vast naar beneden geklapt om ruimte te maken. Aangekomen bij de winkel vraagt Kirsten direct een medewerker naar de tv die ze wil.
… Kirsten voelt
blijdschap
or
boosheid wanneer in de winkel blijkt dat
de tv zelfs in de aanbieding is en ze geld overhoudt
or
de tv die ze wil niet meer op voorraad blijkt te zijn
14
Judith zit op haar fiets voor de supermarkt. Ze is met een groepje van drie vrienden en ze hangen maar wat. Terwijl ze wat dom ouwehoeren ziet Judith een hardloper aankomen. De loper is hevig bezweet en het is hem aan te zien dat hij tot het
Judith stapt snel van haar fiets af en tilt haar fiets aan de kant zodat de loper ongehinderd door kan lopen.
Later die dag loopt Judith van een terrasje door de stad naar huis. Het is nog redelijk vroeg, maar het begint al te schemeren. De zomer loopt ten einde, maar de temperatuur is nog lekker. Judith steekt de
… Judith voelt
woede
or
opluchting als ze zich omdraait en
ziet hoe een man wegrent met haar portemonnee
or
een man haar portemonnee teruggeeft
140
uiterste gaat. Judith en haar vrienden staan pal in het pad van de hardloper. Judith haar vrienden doen een paar stappen opzij om plaats te maken.
brug over en merkt ineens dat haar portemonnee niet in haar zak zit. Ze zou zweren dat ze hem net nog had en draait zich om om te kijken of hij ergens ligt.
15
Adriana loopt door het bos op een zondagochtend. Het is vroeg in de herfst en de bladeren beginnen net te verkleuren. Ze heeft onderweg koffie gehaald bij de enige koffiezaak die op een zondagochtend open is. Terwijl ze wandelt, neemt ze af en toe een slokje. Ze loopt graag in dit bos, het is redelijk groot en ze komt doordeweeks niet veel buiten. Halverwege haar gewone route heeft ze haar koffie op.
Adriana gooit het bekertje in de bosjes en loopt rustig verder, iemand anders raapt het vast wel weer op.
Een uur later loopt Adriana nog steeds in het bos. Ze is ondertussen een heel eind bij de auto vandaan, maar ze is nog lang niet moe. Dan komt haar voet ineens in een gat terecht en verzwikt ze haar enkel. Ze voelt een scherpe pijn en valt om omdat ze niet meer op dat been kan staan. De pijn bonst stug door in haar enkel en ze kan niet anders dan langs het pad blijven zitten.
… Adriana voelt
blijdschap
or
boosheid als er even later iemand langskomt
die haar vriendelijk naar de auto helpt
or
en diegene haar vraag om hulp compleet negeert
16
Kelly loopt door het park op weg naar een feestje. Het schemert al, maar het is nog redelijk licht. Het is een warme dag geweest en er liggen nog mensen op handdoeken op het gras. Kelly pakt haar telefoon om op te zoeken waar ze precies moet
Kelly haalt alle cash eruit, 100 euro, en gooit de portemonnee weer terug op de grond, dat is mooi meegenomen.
Kelly vervolgt haar weg naar het feestje en gaat daarna nog even de kroeg in. De volgende ochtend komt ze redelijk op tijd uit bed. Gelukkig heeft ze geen kater want ze moet nog voor haar studie aan de slag
… Kelly is furieus
or
dolblij als ze in de email leest dat ze
niet mag herkansen en niet kan afstuderen
or
een herkansing mag doen en kan afstuderen
141
zijn. Dan ziet ze vanuit haar ooghoek een portemonnee liggen. Er is niemand te bekennen en Kelly pakt de portemonnee op die vol geld blijkt te zitten.
vandaag. Ze zet de laptop aan en gaat dan eerst koffie maken. Als ze terug komt opent ze haar mail en ziet een mailtje van de studieadviseur.
17
Kim stapt ’s middags in de tram naar huis. Ze checkt in met haar pas en loopt door naar achter om een plaatsje te zoeken. Achter Kim stapt ook nog een vrouw in, ze loopt achter haar aan. Ze is hoogzwanger en volgt Kim verder de tram in. Kim ziet achterin nog één vrije zitplaats en stevent er op af. Voordat Kim de zitplaats bereikt begint de tram schokkend weer te rijden.
Kim aarzelt niet en stapt opzij zodat de zwangere vrouw rustig kan gaan zitten in de slingerende tram.
Eenmaal thuis neemt Kim de stapel post door. Er zit een zoveelste brief bij van de belastingdienst. Ze is al maanden bezig om haar aanslag gecorrigeerd te krijgen. De belastingdienst heeft een fout gemaakt en weigert die te herstellen. Ze hoopt dat het eindelijk is opgelost, ze heeft er genoeg van.
… Kim is opgewekt
or
gefrustreerd
wanneer blijkt dat haar aanslag
nu eindelijk wel klopt
or
nog steeds niet klopt
18
David staat op de steiger op het punt om te beginnen te schuren. Het is halverwege de ochtend en het is al flink heet. Het is de derde week van deze klus en ze zijn met het laatste stuk bezig. Hij is alleen aan het werk aan deze kant, zijn collega's zijn achter bezig. Hij ziet door het raam een vrouw achter
David wil haar niet storen en wacht even tot ze klaar is en waarschuwt dan dat hij lawaai gaat maken.
Na de lunch klimt David weer de steiger op. Hij gaat verder met het schuurwerk, ondertussen is hij bij het laatste raam. Het werk is goed opgeschoten ondanks de hitte. David werkt gestaag door en al gauw is ook het laatste kozijn geschuurd. David zet de
… David is chagrijnig
or
opgelucht als hij ziet hoe het apparaat
van de steiger valt en op de tegels kapot gaat
or
niet van de steiger valt en nog heel is
142
een bureau zitten. Haar telefoon gaat en ze neemt aan en begint een heel gesprek.
machine uit en terwijl hij zijn bezwete voorhoofd afveegt glipt de schuurmachine uit zijn vingers.
19
Michael gaat met zijn vrienden naar een voetbalwedstrijd. Hun club speelt uit tegen de aartsrivaal. De sfeer is gespannen, want de winnaar heeft grote kans kampioen te worden. Ze komen in een vak met medesupporters te zitten. Voor aanvang wordt er om een minuut stilte gevraagd voor een overleden speler van de tegenpartij. Het is rumoerig in het vak en ook Michaels vrienden praten en lachen.
Michael heeft weinig respect voor de tegenstander en zet luidkeels het clublied in, en de rest zingt mee.
Een week later gaat Michael op vakantie. Hij vliegt 's ochtends vroeg naar Cyprus, waar hij een week in de zon wil liggen. Op internet heeft hij een goedkoop appartementje geboekt. De foto's van het appartement waren wat onduidelijk, maar Michael kon de aanbieding niet laten gaan. Eenmaal geland op Cyprus huurt Michael een auto en rijdt naar het appartement.
… Michael is verheugd
or
furieus wanneer bij aankomst blijkt dat het
nog mooier is dan gedacht en het weer schitterend is
or
appartement niet bestaat en hij nergens heen kan
20
Wendy komt wat laat aan bij een feestje van haar broer. Het is een soort housewarming. Haar broer heeft zowel vrienden en familie als collega’s uitgenodigd. Het is een flink groot feest, maar Wendy haar broer had geen zin om drie keer iets te organiseren.
Wendy schudt de man de hand alsof ze het niet doorheeft en groet hem vriendelijk, ze wil de man niet beledigen.
Later op de avond staat Wendy met wat familieleden te kletsen. Een aantal neven en een tante zijn ook aanwezig. Wendy heeft ze al een poos niet gezien en ze praat ze bij over haar leven. Wanneer Wendy even naar de wc gaat en weer
… Wendy is kwaad
or
vrolijk wanneer ze even later
merkt dat ze haar belachelijk maakten
or
ook bij de grap wordt betrokken en meelacht
143
Wendy wordt gespot door haar broer en her en der voorgesteld. Dan stelt haar broer haar voor aan een collega die een mismaakte hand heeft.
terugkomt, treft ze iedereen lachend aan. Ze stoppen abrupt als ze er weer bij komt staan en Wendy kijkt ze vragend aan.
21
Laura komt pas om een uur of acht uit een vergadering. Ze is moe, maar tevreden over de uitkomst. Ze besluit naar huis te lopen omdat het een mooie zomeravond is. Eenmaal uit de drukte van het centrum is het heerlijk rustig op straat. Een paar straten van haar huis ziet Laura het portier van een auto op een kier staan. Ze kijkt rond of er iemand in de buurt is, maar er is niemand.
Laura denkt na en besluit dat ze het beste gewoon het portier kan sluiten zodat niemand de auto leeg kan halen.
De volgende ochtend staat Laura vroeg op en gaat naar haar werk. Ze is nieuwsgierig naar de uitkomst van een subsidieaanvraag die ze heeft ingediend. Bij de vergadering van gisteren heeft ze geprobeerd de laatste twijfelaars over te halen. Het leek erop dat de aanvraag ingewilligd ging worden. Laura probeert gewoon te werken terwijl ze op nieuws wacht.
… Laura is trots
or
woedend wanneer blijkt dat haar aanvraag
is goedgekeurd en haar project kan starten
or
is afgekeurd om een enorm domme reden
22
Leonie gaat naar de schouwburg voor een toneelvoorstelling. Het is druk, want het is een mooi stuk en er spelen veel bekende acteurs mee. Voor de kassa staat een enorme rij, waar Leonie achter aansluit. Plots hoort Leonie iemand verderop
Leonie loopt naar voren en sluit bij haar vriend aan, een vrouw die er wat van zegt scheldt ze uit voor hoer.
Na de voorstelling loopt Leonie naar de garderobe. Ze had een jas en een paraplu meegenomen, omdat er stevige buien waren voorspeld. Bij de garderobe staat personeel dat de jassen en dergelijk ophaalt.
… Leonie voelt
blijdschap
or
irritatie wanneer het garderobepersoneel
haar spullen snel vindt en ze droog thuiskomt
or
pas na tien minuten haar spullen vindt en ze zeiknat wordt
144
haar naam roepen. Verderop in de rij staat een vriend van Leonie naar haar te zwaaien. Hij wenkt Leonie en gebaart dat ze bij hem in de rij moet komen staan.
Leonie haalt het garderobenummer tevoorschijn dat ze had gekregen. Ze geeft het nummer aan het garderobepersoneel, dat op zoek gaat naar haar spullen.
23
Thomas zit in de kroeg met vrienden. Het is een mannenavondje en ze drinken er lustig op los. Thomas heeft vakantie en zijn vriendin is een week bij haar ouders. Hij mist haar wel, maar aan de andere kant is wat vrijheid ook wel fijn. Als iedereen besluit naar huis te gaan blijft Thomas nog even hangen. Hij raakt aan de praat met een mooie vrouw aan de bar die hem wel ziet zitten.
Thomas maakt een praatje met haar en merkt terloops op dat hij een vriendin heeft, en gaat later alleen naar huis.
De volgende dag sleept Thomas zich rond twee uur naar de supermarkt. Hij heeft niks meer in huis en ondanks een lichte kater toch honger. Wanneer hij uit de supermarkt komt zet hij de tassen even neer. Hij heeft gelijk maar grote boodschappen gedaan en het weegt heel wat. Wanneer hij bij zijn flat aankomt, ziet hij iemand vlak voor zich de deur openen.
… Thomas is furieus
or
verheugd als de man omkijkt, even wacht en
dan de deur in zijn gezicht dicht laat vallen
or
de deur vriendelijk glimlachend open houdt
24
Nina staat 's ochtends vroeg op en komt gelijk in actie. Ze gaat vandaag weer naar huis en ze wil nog winkelen voor ze vliegt. Ze heeft in het huis van vrienden van vrienden gezeten. Ze kent ze verder niet, maar ze hebben haar gratis laten logeren.
Nina besluit dat winkelen minder belangrijk is en laat een electricien komen die ze vervolgens zelf betaalt.
Later die dag rent Nina beladen met al haar bagage naar de trein. Ze springt net op tijd in de trein richting het vliegveld. De trein vertrekt op tijd maar tien minuten later staan ze ineens stil tussen de weilanden.
… Nina is opgelucht
or
verbolgen als ze aankomt op het vliegveld en
blijkt dat haar vlucht ook vertraagd is
or
blijkt dat ze te laat is om nog te boarden
145
Nina komt de keuken binnen en trekt de koelkast open. Het lampje springt niet aan en Nina komt er achter dat er een stop is gesprongen.
Nina begint zich zorgen te maken dat ze te laat gaat komen. Een uur later dan gepland komt de trein eindelijk aan op Schiphol.
25
Koen rijdt met de auto door de binnenstad. Het is druk op de weg en er komt een onoverzichtelijke kruising aan. Fietsers en auto's moeten hier de trambaan over vanwege werkzaamheden. Een oudere man op de fiets wankelt onzeker de trambaan op. Het is duidelijk dat de man in de war is en niet goed weet wat hij moet doen. De man komt bijna volledig tot stilstand recht voor Koen zijn auto.
Koen stapt uit om de man veilig naar de overkant te loodsen, terwijl hij de toeterende auto's tot stilte maant.
Drie straten verder slaat Koen rechstaf richting zijn appartement. Hij parkeert een straat verderop waar bijna altijd wel plaats is. Koen zet de motor af en graait zijn tas van de achterbank. Nadat Koen de auto op slot heeft gedaan zwaait hij zijn tas over zijn schouder. Bij de entree van zijn gebouw ziet hij vanuit zijn ooghoek iemand aan komen lopen.
… Koen is uitzinnnig
or
laaiend wanneer de jongen op hem afloopt
en zijn sleutels gevonden blijkt te hebben
or
ineens Koens tas grijpt en hard wegrent
26
Cynthia is op het werk de penningmeester van de personeelsvereniging. Het is een initiatief van en voor werknemers en ze organiseren af en toe gezellige uitjes. Daarnaast verwelkomen ze nieuwe collega's met een cadeautje. Eén keer per jaar innen ze daarvoor vrijwillig contributie van
Cynthia koopt zich die middag een nieuwe winterjas van het geld, er is toch niemand die haar controleert.
De week daarop moet Cynthia naar de dokter. Ze is al een poosje erg vermoeid en heeft weinig eetlust. Ze heeft daarom vorige week bij de dokter bloed laten prikken. De doktersassistente belde gisteren dat ze de volgende ochtend
… Cynthia is kwaad
or
opgelucht als uiteindelijk blijkt dat
de resultaten zoek zijn en ze opnieuw moet
or
de resultaten goed zijn, ze is kerngezond
146
iedereen. Cynthia heeft vorige week een mailtje uitgestuurd met het verzoek te betalen. Ondertussen heeft bijna iedereen het geld overgemaakt.
langs moest komen. Het is standaard procedure dit soort nieuws niet aan de telefoon te vertellen, maar ze is toch nerveus.
27
Frank loopt richting de ingang van het strandpaviljoen. Hij heeft met vrienden afgesproken, maar is aan de vroege kant. Frank loopt blootsvoets over het zand door de branding. Ter hoogte van het paviljoen begint hij aan de klim het strand op. De eerste trede van de trap naar boven is vrij hoog boven het zand. De stap is zo groot dat een jochie van een jaar of vier de trede niet op komt.
Frank hurkt bij het kind neer en helpt hem de trede op en neemt rustig de tijd om hem verder omhoog te helpen.
Frank zoekt binnen een tafel waar iedereen kan zitten en bestelt alvast wat te drinken. Een half uurtje later is iedereen aanwezig en bestellen ze wat te eten. Frank bestelt een biefstuk, medium rare met friet. Wanneer de biefstuk arriveert, blijkt hij bijna well‐done te zijn. Frank roept de ober terug om het probleem uit te leggen en laat de biefstuk zien.
… Frank is ziedend
or
blij als de ober reageert op een
boze manier en net doet of hij zeurt
or
vriendelijke en correcte manier
28
Remco zit in de auto, op weg naar zijn moeder. Hij is de stad nog niet uit en het spoelt van de regen. Het regent onophoudelijk sinds gistermiddag en er hebben zich flinke plassen gevormd. Op sommige plekken staat de straat volledig blank. Er zijn weinig auto's
Remco geeft extra gas om een zo groot mogelijke golf water te creëren, de wandelaar wordt zeiknat gespetterd.
Eenmaal buiten de stad schiet het lekker op op de snelweg. Het blijft rustig op de weg en Remco geniet van het landschap en de rit. Hij heeft de radio hard aan staan en zingt luidkeels mee. Wanneer hij naar het dashboard kijkt om de
… Remco is gefrustreerd
or
verheugd als hij na een paar minuten rijden
geen bezine meer heeft en stil komt te staan
or
een bezinestation ziet en hij kan tanken
147
en al helemaal weinig fietsers en voetgangers op straat. Remco stevent op een gigantische plas af en ziet iemand op de stoep lopen.
zender bij te stellen ziet hij een lampje. Hij is vergeten te tanken en de benzine was al heel lang echt bijna op.
29
Maarten is als enige van zijn team op werkbezoek in het buitenland. De bedoeling is om een kijkje te nemen bij een vergelijkbaar team op het hoofdkantoor. Maarten weet dat dit team op hun projecten uit is. Ze willen maar wat graag dat Maartens afdeling moet sluiten. De teamleider van het rivaliserende team wil een één op één gesprek met Maarten. Hij nodigt Maarten uit om bij hen te komen werken als hij zijn team saboteert.
Maarten denkt alleen maar aan zichzelf en stemt meteen toe, hij hoort toch liever bij het winnende team.
Later die week vliegt Maarten terug naar huis. Het is een lange vlucht en hij heeft onderweg veel om over na te denken. Als het vliegtuig eindelijk landt wacht hij ongeduldig op zijn bagage. Het is hartstikke vroeg en hij wil zijn vrouw verrassen met een ontbijt op bed. Onderweg naar huis koopt hij allerlei lekkere dingen en gaat dan snel door naar huis.
… Maarten voelt
razernij
or
blijdschap als hij eenmaal thuiskomt
en zijn vrouw met ander in bed aantreft
or
en ze samen een romantische ochtend hebben
30
Jordy loopt over het strand terug naar zijn handdoek. Onderweg komt hij langs een kraampje dat friet en dergelijke verkoopt. Het is er niet druk, het is überhaupt niet druk vandaag. Ondanks dat het een mooie dag is, is er maar een tiental mensen.
Jordy blijft staan, lacht haar vierkant uit om haar gestuntel en loopt dan nog steeds geamuseerd door.
Jordy gaat nog een stukje zwemmen in de aanzienlijke golfslag. Daarna gaat hij liggen drogen in de zon en leest zijn boek. Hij gaat helemaal op in het verhaal en de tijd verstrijkt. Op een gegeven
… Jordy is pissig
or
blij als de buschauffeur hem ziet
en gewoon doorrijdt en hem daar laat staan
or
en vriendelijk even wacht zodat hij mee kan
148
Jordy loopt voorbij een meisje met een mitella dat probeert een flesje cola te openen. Het wil niet echt lukken met één hand en ze kijkt gefrustreerd.
moment bedenkt hij zich dat de laatste bus vrij vroeg gaat op zondag. Hij checkt hoe laat het is en bedenkt zich dat hij zal moeten opschieten om de bus te halen.
31
Ruben zit op de bank met zijn laptop op schoot wanneer de bel gaat. Hij zet de laptop op de bank naast zich en staat op. Het is een pakketje voor iemand anders in het gebouw. Ruben neemt het aan en ziet op de pakbon staan dat het een geluidsinstallatie is. Ruben gaat weer verder met waar hij mee bezig was. Later die dag gaat de deurbel en het is de buurvrouw die haar pakketje komt halen.
Ruben doet net of hij geen pakje heeft en houdt stug vol tot ze weg gaat, de stereo zet hij in zijn slaapkamer.
De volgende dag is Ruben onderweg naar huis van de supermarkt. Hij heeft twee volle en zware tassen boodschappen. Bij de stoplichten zet hij de tassen naast zich neer en wacht op groen. Wanneer het groen wordt pakt hij de tassen op en haast zich naar de overkant. Bijna aan de overkant struikelt Ruben over de stoeprand en valt tegen de man voor hem aan.
… Ruben wordt
witheet
or
vrolijk wanneer de man voor hem
hem een duw geeft en hij dus keihard valt
or
hem vriendelijk helpt overeind te blijven
32
Dennis stapt de lift in in het ziekenhuis. De vrouw van zijn broer is gisteren bevallen en hij gaat op kraamvisite. Hij heeft een grote knuffel onder zijn arm en een bos bloemen. Hij stapt op de begane grond in en moet op de vijfde verdieping
Dennis groet haar vriendelijk en maakt een praatje met haar voordat hij op de vijfde verdieping uitstapt.
Wanneer hij eenmaal zijn jongste neefje heeft bekeken gaat Dennis weer. Hij is met de bus, die stopt vlak bij hem voor de deur. Hij is binnen een half uurtje thuis en maakt wat te eten. In de keuken ziet hij op het bord zijn to‐do
… Dennis is kwaad
or
dankbaar als hij de tuin inkijkt en ziet dat
de buren hun tuinafval in zijn tuin hebben gegooid
or
de buurman al het werk voor hem heeft gedaan
149
zijn. Op de derde verdieping stopt de lift en stapt een vrouw in. De vrouw is kaal en mager, ze heeft wallen onder haar ogen en een grauwe huid.
lijstje en herinnert zich dat de tuin een wildernis is. Hij werpt een snelle blik naar buiten om te zien hoe erg het is.
33
Saskia is vrijwilliger bij een grote manege en vandaag hebben ze open dag. Vanuit de hele omgeving komen ouders en vooral kinderen naar de manege. Er worden rijlessen en workshops over paardenverzorging gegeven. Saskia heeft de hele ochtend workshops gegeven.‘s Middags laat ze jonge kinderen ritjes maken in de bak. Wanneer Saskia zich omdraait naar het volgende kind ziet ze meteen dat het downsyndroom heeft.
Saskia doet extra haar best om het meisje een mooie tijd te geven en ze straalt van geluk op het paard.
De volgende ochtend is Saskia vroeg op haar normale werk. Ze werkt bij een groot accountantsbedrijf. Ze moet om 9:00 bij de baas komen en ze weet niet precies waarvoor. Ze heeft er slecht van geslapen vannacht terwijl ze probeerde te verzinnen waar het over zal gaan. Wanneer het eenmaal negen uur is en Saskia bij de baas zit, steekt hij gelijk van wal.
… Saskia is laaiend
or
uitzinnig wanneer de baas in het gesprek
haar onterecht dingen aanrekent en ontslaat
or
vertelt dat ze promotie krijgt en opslag
34
Tim loopt 's middags door de supermarkt. Op zijn lijstje staat van alles om zelf een verjaardagstaart te maken. Wanneer hij alles heeft gevonden gaat hij naar de kassa. Voor hem in het gangpad loopt een oudere vrouw. Ze is duidelijk ook
Tim houdt netjes zijn pas in en laat de oude dame voor gaan in de rij, hij heeft tenslotte geen haast.
Later die dag is hij bezig de taart te bakken, hij staat in de oven. Hij viert vandaag zijn verjaardag en heeft iedereen eigengemaakte taart beloofd. Tim staat niet bepaald bekend om zijn
… Tim is blij
or
geïrriteerd wanneer de cake uiteindelijk
perfect uit de oven komt
or
aangebrand uit de oven komt
150
onderweg naar de kassa, maar ze loopt minder snel dan Tim. Het gangpad is niet zo heel breed en er komt iemand van de andere kant aanlopen.
culinaire hoogstandjes. Een aantal van zijn vrienden en familie reageerde dan ook nogal sceptisch. Het is nu een kwestie van eer geworden dat deze taart goed lukt.
35
Joris loopt 's ochtend over de markt richting station. Hij heeft zijn oordopjes in en luistert naar muziek. De marktkooplui zijn in de weer met het opbouwen van hun kramen. De klinkers zijn wit uitgeslagen van de vorst van vannacht. Een man passeert voor Joris met een stapel kisten vol mandarijnen in zijn armen. De bovenste wankelt wat en even later ligt de straat bezaaid met mandarijnen.
Joris stopt en helpt de man gauw met het oprapen van de mandarijnen en maakt ondertussen een praatje.
Later die ochtend komt Joris terug van de lunch. Hij gaat weer achter zijn bureau zitten en kijkt naar zijn lijstje met taken voor vandaag. Hij is lekker opgeschoten vanochtend, maar kan nu eigenlijk niet verder. Hij wacht op bericht van een afdeling in het buitenland. Hij heeft ze gemaild of ze op willen schieten omdat hij niet verder kan zonder de info.
… Joris is woest
or
blij als hij een mailtje krijgt met daarin
informatie die hij 3 maanden geleden nodig had
or
eindelijk het verlossende woord
36
Mark rent snel terug richting het eigen doel na de gefaalde aanval. De tegenstander heeft net de bal terugveroverd. Zijn team heeft moeite met terugschakelen naar verdedigen. Over de flank krijgt de spits van de
Mark trekt zijn been gauw in omdat hij niemand wil verwonden, ook al kan daardoor de aanval wel doorgaan.
De bal komt bij de keeper terecht en die speelt Mark aan. Mark neemt de bal aan en begint een soloactie langs de lijn. Hij passeert twee man en een derde komt veel te laat aanrennen. Mark trekt naar binnen
… Mark is trots
or
boos als hij aanlegt om te schieten
en de keeper passeert en scoort
or
en hij keihard onderuit wordt gehaald
151
tegenstander een geweldige lange pass. Mark sprint om hem af te snijden op weg naar het doel. Mark zet een tackle in en beseft zich meteen dat hij veel te hard in heeft gezet.
richting de zestien meter van de tegenstander. Zijn teamgenoten zijn nog niet mee naar voren en dus legt Mark zelf aan om uit te halen.
37
Myrthe krijgt haar schoonouders vanavond op bezoek en ze blijven eten. Ze wil indruk op hen maken met haar kookkunsten. Ze heeft alleen nog niks in huis en rijdt daarom naar de supermarkt. Myrthe parkeert haar auto op de parkeerplaats en loopt naar de ingang van de supermarkt. Bij de ingang staat een man die overduidelijk dakloos is. De man loopt op Myrthe af en vraagt om wat kleingeld.
Myrthe spuugt naar de man, geeft hem een duw, en sist hem toe 'kruip terug in een hoekje vieze mongool'.
Wanneer Myrthe de boodschappen heeft gekocht, rijdt ze weer naar huis. Thuis aangekomen loopt ze met haar boodschappen naar de voordeur. Myrthe kan haar voordeursleutel niet vinden. Op dat moment bedenkt ze zich dat ze zonder sleutel de deur achter zich dicht heeft gedaan. Snel loopt ze naar het huis van haar buren, die een reservesleutel hebben.
… Myrthe is opgelucht
or
getergd als ze bij de buren aanbelt en
die thuis blijken te zijn in ze dus haar huis in kan
or
die op vakantie zijn en ze een slotenmaker moet bellen
38
Bram is op stap met een groepje vrienden. Ze kennen elkaar nog van hun studietijd en spreken af en toe af. Bram heeft al een aantal biertjes op en de rest heeft hem ook aardig zitten. Op een gegeven moment besluiten ze naar de volgende kroeg te
Bram blijft kalm en springt er tussen, hij dwingt zijn vriend excuus te maken voordat de boel kan escaleren.
Daarna is de avond snel afgelopen en gaat Bram naar zijn fiets. Zijn fiets staat helemaal aan de andere kant van het centrum. Het is ondertussen drie uur geweest en hij wil niets liever dan slapen. Het schiet Bram
… Bram is gelukkig
or
razend als hij aankomt waar hij zijn fiets had staan
en ziet dat deze er gewoon nog staat
or
en hij echt gejat is en hij naar huis kan lopen
152
gaan. Voor de ingang van de kroeg staat een groepje kerels te roken. Één van Bram zijn vrienden loopt op ze af en begint ruzie te zoeken.
ineens te binnen dat hij zijn fiets helemaal niet op slot heeft gezet. Hij graaft in zijn zakken en kan de sleutel nergens vinden.
39
Kevin loopt 's avonds naar het huis van zijn ouders. Hij is bij een vriend geweest die hij niet had gezien sinds hij is verhuisd. Het is schemerig en het is een eind lopen, maar dat vindt Kevin niet erg. Het is fijn om weer door bekende straten te slenteren. Verderop ziet Kevin een gedaante op vier poten uit de bosjes komen. Het is een gewonde hond die zacht jankt en naar hem toe loopt.
Kevin belt gauw de dierenambulance en aait het beestje en stelt hem gerust terwijl hij op de ambulance wacht.
Eenmaal thuis na het incident met de hond gaat Kevin direct naar bed. De volgende dag staat hij vroeg op om nog wat mensen op te zoeken. Hij leent de auto van zijn ouders omdat sommigen buiten de stad wonen. Halverwege, midden tussen de weilanden, houdt de motor er mee op. Na 20 minuten wachten langs de weg komt er eindelijk een andere auto aan.
… Kevin is uitgelaten
or
pissig wanneer de andere automobilist
stopt en zijn auto kan repareren
or
wel langzamer gaat rijden, maar niet stopt
40
Anouk zit alvast in de vergaderruimte waar ze straks overleg heeft. Ze neemt de documenten die ze heeft voorbereid nog eens door. In de excel sheets van de begroting spot ze ineens een grove fout. Deze fout betekent dat het project veel meer gaat kosten dan ze had gezegd. Anouk heeft weinig zin
Anouk besluit de schuld te geven aan een medewerker onder haar die waarschijnlijk toch ontslagen gaat worden.
De vergadering verloopt verder prima en Anouk gaat door met haar dag. Tegen een uur of acht is ze thuis en haalt de post op. Eenmaal binnen ziet ze een brief van het elektriciteitsbedrijf. Het is de jaarafrekening en dat was vorig jaar een dure grap. Anouk heeft
… Anouk is driftig
or
opgetogen
als uit de brief blijkt dat ze
wederom een groot bedrag moet bijbetalen
or
dit jaar zelf een fiks bedrag terugkrijgt
153
om dit op te moeten biechten. Voordat ze echt na heeft kunnen denken komen haar bazen binnen.
toen het voorschot bij laten stellen en hopelijk hoeft ze nu niet bij te betalen.
41
Jesse komt 's middags rond half vier aanrijden bij de school. Hij parkeert in de rij wachtende ouders. Er zijn nog geen kinderen op het plein en hij wacht rustig. Hij is zo verzonken in gedachten dat hij niet merkt dat de kinderen naar buiten komen. Hij is dan ook verrast wanneer zijn dochtertje plots achterin klimt. Ze heeft een tekening gemaakt die ze vol trots laat zien en aan hem wil geven.
Jesse wijst haar erop dat ze nog steeds slordig en buiten de lijntjes kleurt en dat ze beter haar best moet doen.
Later die avond als zijn dochtertje al op bed ligt kijkt Jesse voetbal. Hij heeft geld gezet op de uitkomst van deze wedstrijd. Het was een flink bedrag en hij wil graag winnen, want hij kon het geld eigenlijk niet missen. Vlak voor tijd staat het 1‐1 en hij heeft geld gezet op 2‐1 winst. Hij kijkt gespannen toe hoe zijn team een laatste aanval inzet.
… Jesse is woest
or
uitgelaten wanneer vlak voor het einde
de tegenpartij scoort en hij zijn geld kwijt is
or
zijn team scoort en hij dus 300 euro wint
42
Inge zit 's middags in de trein op weg naar Groningen. Het is ontzettend rustig in de trein sinds Zwolle. Er zitten naast Inge maar vier mensen in de hele coupé. Er zit een man in de vierzits aan de overzijde van het gangpad een boek te lezen. Wanneer de trein weer
Inge springt gauw overeind en weet de man nog net voordat de deuren sluiten zijn tas te overhandigen.
Eenmaal in Groningen aangekomen pakt Inge de bus richting Oranjewijk. Ze logeert bij een vriendin en ze gaan naar een concert in de Oosterpoort vandaag. De bus doet er iets langer over dan normaal, maar Inge heeft geen haast. Ze stapt uit en
… Inge voelt
vreugde
or
woede wanneer ze te horen krijgt
dat haar bod op een huis is geaccepteerd
or
dat er bij haar is ingebroken en alles weg is
154
stopt staat de man op om uit te stappen. Ineens ziet Inge dat de man een tas onder de bank heeft laten staan.
loopt de laatste vijf minuten naar haar huis. Vlak voor ze aanbelt, gaat haar telefoon en ze neemt op.
43
Julia zit op haar plek achter de receptie, de telefoon staat roodgloeiend. Ze handelt het ene na het andere telefoontje af. Voordat Julia het weet is de ochtend alweer voorbij. Na de lunch is het iets minder druk. Dan gaat de telefoon weer, het is iemand voor de man van accounting. De accountant heeft de laatste tijd steken laten vallen en ook nu belt er iemand met een probleem.
Julia geeft de boodschap lekker niet door en geniet wanneer de man daardoor in problemen komt en ontslagen wordt.
Later die week is Julia aan het winkelen. Ze zoekt een specifiek computerspelletje voor de verjaardag van haar broertje. Ze heeft hem beloofd dat hij deze morgen voor zijn verjaardag krijgt. Julia heeft echter weinig tijd gehad om te winkelen de laatste tijd. Eindelijk vind ze één kopie van het spelletje, maar een andere klant ziet hem tegelijkertijd.
… Julia is geïrriteerd
or
vrolijk als de andere klant het spelletje
snel afrekent en haar compleet negeert
or
aan haar geeft zodat ze haar cadeau heeft
44
Larissa is lerares op een basisschool. Ze geeft les aan de bovenbouw, groep vijf en zes. Op een woensdagochtend heeft ze pleinwacht tijdens de pauze. Ze loopt grote rondes over het plein en houdt alles goed in de gaten. Af en toe spreekt ze een leerling aan, maar verder
Larissa wil ze tot de orde roepen, maar ziet dan dat het een moslimvrouw is en laat ze lekker hun gang gaan.
Later die week gaat Larissa naar de bank voor een afspraak. Ze wil haar hypotheek oversluiten en snapt niet veel van financiën, dus wil ze persoonlijk advies. Ze loopt de bank binnen en gaat naar de balie om zich te melden. De receptionist
… Larissa is gerustgesteld
or
laaiend als ze twee uur later naar buiten loopt
met een nieuwe en zeer gunstige hypotheek
or
zonder resultaat omdat ze haar afspraak waren vergeten
155
is het rustig. Dan ziet ze een stel leerlingen uit groep 8 aan de rand van het plein rommel staan gooien naar een vrouw die langsloopt.
e zegt haar plaats te nemen, ze komen haar zo halen. Larissa gaat zitten en leest een tijdschrift terwijl ze rustig wacht.
45
Hendrik zit 's ochtends al vroeg op het kantoor. Hij is druk bezig met het tentamen schrijven dat zijn studenten volgende week moeten maken. Terwijl hij daar mee bezig is wordt er op de deur geklopt. Het is één van de studentes uit zijn werkgroep, een knappe verschijning. Het is al gauw duidelijk dat ze probeert Hendrik te verleiden om haar het tentamen in te laten zien. Ze zinspeelt op hoe dankbaar ze zou zijn.
Hendrik maakt misbruik van de situatie en belooft haar het tentamen als ze vanavond bij hem thuis komt.
Later die dag moet Hendrik college geven in een gebouw buiten de binnenstad. Hij pakt de fiets daarheen en trapt stevig door. Op een gegeven moment fietst hij op een drukke kruising af en stopt hij voor het licht. Even later springt het op groen en Hendrik steekt over. Halverwege de kruising vliegt de ketting er ineens af en staat hij midden op de kruising stil.
… Hendrik is dankbaar
or
getergd als de auto's die nu groen hebben
rustig wachten en hem vriendelijk de tijd geven
or
ongeduldig beginnen te toeteren en optrekken
46
Eva loopt snel over het plein richting het station. Ze gaat vandaag naar Haarlem voor een sollicitatiegesprek. Ze heeft zich goed voorbereid en is ruim op tijd van huis gegaan. Vlakbij de ingang naar de stationshal staat een zwerver te bedelen.
Eva vist wat muntgeld op en gooit dit in het bekertje van de zwerver, ze geeft altijd wat ze kan missen.
Ze loopt snel door en kijkt op het bord welk spoor ze moet hebben. Ze ziet dat er over 2 minuten een trein gaat en zet het op een rennen. Als ze deze trein kan halen, dan komt ze sowieso op tijd. Het is druk in de tunnel naar de sporen
… Eva is dankbaar
or
woedend als de conducteur haar ziet rennen
en vriendelijk even op haar wacht
or
de deuren expres voor haar neus sluit
156
Wanneer Eva langsloopt, houdt hij een papieren bekertje omhoog met wat kleingeld erin. De man kijkt hoopvol hoe Eva in haar broekzak graaft.
en Eva moet steeds uitwijken. Eindelijk bereikt ze de trap en sprint met drie treden per keer naar boven.
47
Wouter werkt op de marketingafdeling van een groot bedrijf. Op dezelfde afdeling werkt zijn beste vriend. Ze werken al jaren heel veel samen en ze zijn een hecht team. Op een ochtend wordt Wouters vriend naar het kantoor van de baas geroepen. Wouter weet niet waar het over gaat, maar hij is wel nieuwsgierig. Later die dag hoort Wouter dat zijn vriend een promotie heeft gekregen en hij niet.
Wouter besluit de vriendschap te beëindigen en verspreidt een roddel dat hij geld zou hebben gestolen.
Later die week heeft Wouter een zeldzame vrije dag. Hij besluit de meubelzaken bij langs te gaan voor nieuwe meubels. Bij IKEA ziet hij een leuke garderobekast, die perfect past in zijn slaapkamer. Wouter aarzelt geen moment en koopt de kast meteen en gaat weer richting huis. Wouter begint thuis meteen enthousiast het bouwpakket in elkaar te zetten.
… Wouter is tevreden
or
woedend als bij het monteren blijkt
dat de kast perfect past waar hij hem wilde
or
dat de planken scheef zijn en er dingen ontbreken
48
Elise zit in de bus op weg naar de film. Ze heeft haar oordopjes in en zoekt de filmtijden nog eens op. De bus is halverwege en ze sms't haar vrienden dat ze eraan komt. Elise verandert de muziek naar iets stevigers en zet het harder. Er zitten weinig mensen in de bus behalve
Elise verontschuldigt zich en zet haar muziek meteen zachter, vervolgens wenst ze de dames nog een fijne dag.
Twee haltes verder stapt Elise uit en loopt richting bioscoop. Bij de ingang ontmoet ze haar vrienden en ze gaan gauw naar de zaal. Na de film gaan ze nog even gezellig wat drinken. Elise bestelt een biertje, maar bedenkt zich later en loopt naar
… Elise wordt
chagrijnig
or
vrolijk als de serveerster reageert
op een norse manier en haar boos aankijkt
or
op een vriendelijke manier en ze kletsen wat
157
twee oudere dames voor haar. Eén van hen draait zich om om te zeggen dat haar muziek wel heel hard staat.
de bar om iets anders te bestellen. Ze zegt tegen de serveerster dat ze toch liever een witte wijn heeft.
49
Tom loopt naar huis van de supermarkt aan de overkant van de straat. Hij heeft één tas met boodschappen, maar die weegt niet zoveel. Met zijn vrije hand opent hij de zware portiekdeur van de flat. Achter hem loopt een vrouw met twee tassen, ze heeft moeite met het gewicht. Tom herkent haar en weet dat ze ook de flat in moet. Hij stopt even in de deuropening en denkt na wat te doen.
Hij wacht rustig tot ze er is en houdt de deur vriendelijk voor haar open, wat ze zichtbaar op prijs stelt.
Later die dag wil Tom een afspraak maken bij de dokter. Hij zoekt het nummer op en pakt de telefoon. Hij wil al een hele tijd een afspraak maken vanwege een zere knie. Tom is echter zo druk dat hij vaak geen zin heeft om te bellen, het wachten duurt altijd zo lang. Hij moet eigenlijk al weer de deur uit, maar wil toch even snel bellen.
… Tom is verrast
or
pissig wanneer de telefoon over gaat
en hij gelijk de assistente krijgt
or
en hij de 15de wachtende blijkt te zijn
50
Niels is 's ochtends vroeg als eerste op kantoor. Het is verder muisstil op de afdeling. Na een half uurtje heeft Niels al flink wat werk verzet, dit is zijn favoriete deel van de dag. Hij legt de laatste hand aan een document en streept hem van zijn lijstje. Vervolgens gaat hij koffie halen en komt de eerste
Niels zegt goedemorgen en merkt op dat het dieet niet lijkt te helpen, misschien minder suiker in de koffie?
Niels werkt rustig door tot de lunch en daarna heeft hij een vergadering. Zijn team presenteert vandaag de resultaten van een project. Het project is een enorm succes geworden voor het bedrijf. Niels presenteert vandaag zelf niet, maar hij weet dat hij een integraal onderdeel
… Niels voelt
razernij
or
blijdschap wanneer zijn manager in de meeting
liegt en alle eer voorzichzelf opstrijkt
or
hem prijst voor zijn originaliteit en inzet
158
collega tegen. Niels kent haar vaag en weet toevallig dat ze sinds kort op dieet is.
van dit succes is. De belangrijkste ideeën voor het nieuwe programma kwamen van hem.
51
Mike is op weg naar de sportschool op een doordeweekse avond. Het begint net wat te schemeren en het regent gestaag. Hij steekt over bij de stoplichten en gaat de hoek om. Aan het eind van de straat ziet hij mensen in en uit lopen bij de sportschool. Aan de overkant van de straat fietst een oudere vrouw de andere kant op. Mike ziet hoe ze plots begint te slingeren en met fiets en al omvalt.
Mike bedenkt zich niet en rent er naar toe en helpt haar overeind, ze mankeert niks maar is blij met de hulp.
Eenmaal in de sportschool kleedt Mike zich om. Dan loopt hij snel naar de loopbanden, waar het altijd druk is. De loopbanden staan boven en halverwege de trap ziet hij dat zijn veter los zit. Hij bukt zich om deze te strikken en haast zich dan verder naar boven. Hij ziet dat er nog eentje vrij is, maar een andere man loopt ook op de band af.
… Mike is dankbaar
or
gepikeerd wanneer de man die ook wil rennen
hem vriendelijk voor laat gaan
or
net voor hem op de band springt
52
Rianne loopt door het bos met de hond. Het is zaterdagmiddag en er schijnt een waterig zonnetje. Terwijl de hond druk bezig is met snuffelen belt Rianne met een vriendin. Ze hebben elkaar al een poosje niet gesproken en praten bij. Rianne doet een verhaal uit de doeken over een dronken avond op zakenreis.
Rianne gaat gelijk zachter praten en hangt snel op, ze knikt de rouwers toe bij wijze van groet en excuus.
Later die dag bedenkt Rianne zich dat ze wel zin heeft om wat te doen die avond. Ze is al een hele poos niet meer op stap geweest. Ze stuurt een berichtje op de groepsapp van een groep vrienden van het volleyballen. Ondertussen gaat Rianne
… Rianne wordt
woest
or
vrolijk als ze leest dat iedereen
zegt niet te kunnen terwijl ze weet dat ze wel uitgaan
or
enthousiast is en ze een leuke avond heeft
159
Rianne merkt ineens dat ze langs het kerkhof loopt en dat een groep rouwers bij een nieuw graf staat.
alvast iets te eten klaar maken. Tijdens het eten checkt ze haar telefoon en leest de reacties van iedereen.
53
Bart stapt uit de auto en loopt richting het uitvaartcentrum. Hij is hier met een paar andere mensen van het werk voor de uitvaart van een collega. Hij kende de man niet goed, maar sommigen van zijn collega's waren wel close. Hij zoekt samen met zijn collega's een plekje achterin. De dienst begint en na het eerste nummer spreekt zijn vrouw. Een collega van Bart begint zachtjes te huilen.
Bart slaat een arm om haar heen om haar te troosten en haar steun te bieden, wat ze duidelijk op prijs stelt.
Een week later heeft Bart een belangrijke afspraak met zijn manager.Het is een vervroegd kwartaalgesprek om zijn functioneren te evalueren. Bart heeft geen redenen om te denken dat zijn manager iets te klagen zou hebben. Maar zo'n gesprek is toch altijd spannend en stressvol. Als ze eenmaal zitten en de beleefdheden hebben gehad, steekt zijn manager van wal.
… Bart is opgelucht
or
furieus wanneer zijn manager hem vertelt
dat hij een promotie en opslag krijgt
or
dat hij per direct ontslag krijgt
54
Sander werkt bij een groot softwarebedrijf. De directeur van het bedrijf is kort geleden met pensioen gegaan. Sander zit in de sollicitatiecommissie om een geschikte nieuwe directeur te vinden. De eerste brieven en cv's van kandidaten
Sander denkt dat het goed zou zijn om een vrouw aan het roer te hebben, en hij zet de vrouwelijke kandidaten gewoon op de lijst.
Wanneer Sander klaar is met werken rijdt hij het half uurtje naar huis. Eenmaal thuis besluit hij een eindje te gaan hardlopen. Hij trekt zijn hardloopkleding aan en loopt zijn huis uit. Na wat warming‐up en rekken en strekken gaat hij op
… Sander wordt
boos
or
blij als het groepje jongeren
expres de weg blijft blokkeren en lacht
or
gauw aan de kant gaat zodat hij door kan lopen
160
zijn binnengekomen. Sander print de documenten uit en maakt een overzicht van alle kandidaten. Het valt hem op dat er veel vrouwelijke kandidaten zijn.
pad. Na een half uurtje stevig lopen komt Sander bij een smalle straat waar een groepje jongeren de weg blokkeert.
55
Tamara is eigenaar van een koffiezaak. Ze heeft vandaag een aantal sollicitatiegesprekken. Het gaat voornamelijk om tijdelijke krachten die in de vakantieperiode bijspringen. Toch moet ze ook die zorgvuldig selecteren. De eerste drie kandidaten zijn niet super geschikt op het eerste gezicht. Tamara is verbaasd als de volgende een enigszins ruig uitziende Turkse jongen blijkt te zijn.
Tamara interviewt de jongen en uiteindelijk blijkt hij de meest geschikte kandidaat en krijgt de baan.
Aan het einde van de dag is Tamara met het schoonmaakwerk bezig. De balie is schoon en de vloer is geveegd, maar nog niet gedweild. Er zit nog één klant in de hoek en ze zijn officieel nog open. Tamara wacht daarom met dweilen en vult de suikerpotten bij. Tamara loopt met een zwaar dienblad vol suikerpotten als de laatste klant vol tegen haar op botst.
… Tamara wordt
furieus
or
blij als de klant die haar aanstootte
geen sorry zegt en zelfs geamuseerd omkijkt
or
aardig blijkt te zijn en helpt op te ruimen
56
Thijs stapt snel op de fiets om nog iets te halen in de stad. Hij fietst de straat uit en richting centrum. Hij zet zijn fiets op slot tegen een paal bij het gemeentehuis. Hij hoeft maar één ding te halen en neemt verder niet de tijd om rond te kijken. Hij
Thijs duwt haar zowat omver wanneer hij zich langs haar wurmt, 'aan de kant, lelijke mongool' bijt hij haar toe.
Thijs vervolgt zijn weg naar de winkel en koopt snel wat hij nodig heeft. Daarna baant hij zich zo snel mogelijk een weg terug door de binnenstad. Halverwege voelt hij ineens een zachte aanraking. Hij voelt in
… Thijs is ziedend
or
blij als hij ziet hoe de omstanders
domweg opzij gaan en de dief ontsnapt
or
de dief grijpen voordat deze weg kan komen
161
slalomt door de mensenmassa heen en schiet lekker op. Ineens staat hij vlak voor een gehandicapte jonge vrouw met een rollator.
zijn zak, merkt dat zijn portemonnee weg is en ziet ineens iemand snel weglopen. Thijs schreeuwt dat hij bestolen is en de dief zet het op een hollen.
57
Jeroen is twee weken op vakantie in Italië. Hij heeft een groot deel van de eerste week aan het strand doorgebracht. De tweede week wil hij iets meer cultuur opsnuiven. Vandaag heeft hij de trein genomen naar een oud stadje vlakbij. Er schijnt een heel mooie en oude kerk te staan. Wanneer hij in zijn hemd de kerk wil binnenlopen, wijst iemand hem op een bord met kledingvoorschriften.
Jeroen verontschuldigt zich uitvoerig en haalt een shirt met lange mouwen uit zijn tas en trekt dit snel aan.
Jeroen gaat op het terras zitten bij het enige restaurant dat hij heeft gezien. Hij spreekt geen Italiaans en hoopt dat ze Engels spreken. Hij probeert uit te puzzelen wat er op de kaart staat, hij is vegetariër. Het is een warme dag en Jeroen is blij met de schaduw terwijl hij op de ober wacht. Als de ober langskomt, vraagt Jeroen of hij Engels spreekt.
… Jeroen is witheet
or
vergenoegd
als de ober vervolgens
zonder wat te zeggen omdraait en hem negeert
or
geduldig zijn best doet om Engels te spreken
58
Patrick heeft sinds een jaar een eigen bedrijf. Het bedrijf is zo gegroeid dat Patrick een secretaresse nodig heeft. Hij heeft heel veel reacties gehad. Eén sollicitante steekt boven alle anderen uit door haar ervaring en opleiding. Patrick nodigt haar uit en is tijdens het gesprek erg enthousiast.
Patrick denkt nog eens na en besluit inderdaad dat hij liever een lekker ding heeft en neemt een ander aan.
Later die dag werkt Patrick aan een verslag voor een klant van zijn bedrijf. Het is een eindrapportage van alle werkzaamheden. Het is een groot document geworden dat Patrick veel kostbare tijd heeft gekost. Wanneer Patrick de laatste punt neerzet,
… Patrick is opgelucht
or
ziedend wanneer na het opnieuw opstarten blijkt
dat er een automatische back‐up beschikbaar is
or
dat het document verloren is en het opnieuw moet
162
Zijn zakenpartner wijst hem er later op dat ze wel wat ouder is en andere kandidaten er leuker uitzagen.
valt ineens de stroom uit. Patrick bedenkt zich dat hij tussendoor niet heeft gesaved en start de computer weer op.
59
Anna staat aan de kassa bij de supermarkt. Terwijl de kassière de laatste boodschappen scant, pakt ze zoveel mogelijk in. Nadat ze het laatste artikel heeft gescand noemt ze het bedrag. Anna betaalt met 50 euro en de kassière geeft haar wisselgeld terug. Anna ziet meteen dat ze haar veel te veel wisselgeld geeft. Ze weet dat kassières flink problemen kunnen krijgen bij een kasverschil.
Anna wijst haar er dan ook vriendelijk op dat ze een fout heeft gemaakt, ze wil niet dat ze ontslagen wordt.
Na de supermarkt gaat ze ook nog even naar de boekwinkel. Ze moet echter opschieten want haar parkeerkaartje verloopt zo. Bij de boekwinkel kan ze niet gelijk vinden wat ze zoekt. Ze vraagt het de eigenaar en die helpt haar het boek te vinden. Anna rekent af en loopt, zo snel als ze kan met alle tassen, naar de parkeerplaats, ze is nu echt te laat.
… Anna wordt
vrolijk
or
kwaad als de parkeerwachter haar ziet
en besluit de boete te verscheuren
or
en toch alsnog de boete uitschrijft
60
Nienke loopt door het park op een zondagochtend. Ze is op weg naar de winkel om brood te halen. In de verte ziet ze een oud mannetje met een rollator lopen. Nienke loopt rustig door, maar heeft de man uiteraard al gauw bijna ingehaald. De man stopt even en graaft een zakdoek uit zijn
Nienke roept de man snel na dat hij iets verliest en raapt het voor hem op zodat de man niet hoeft te bukken.
Later die week loopt Nienke met een laptop onder de arm. Ze is onderweg naar de winkel waar ze hem de dag daarvoor heeft gekocht. Toen ze hem gisteravond probeerde te installeren bleek hij het niet te doen. Ze is helaas het bonnetje
… Nienke is gepikeerd
or
gelukkig als de man vervolgens zegt dat
ze zonder bonnetje naar haar geld kan fluiten
or
ze ook zonder bonnetje haar geld terugkrijgt
163
broekzak. Nienke ziet hoe tegelijk met de zakdoek ook een pasje op de grond valt, maar de man merkt niks.
kwijt, maar hoopt dat ze haar nog herkennen. De man in de winkel herkent haar inderdaad en vraagt om het bonnetje.
61
Marloes is alweer een tijdje vrijgezel. Sinds kort is er bij Marloes op het werk een jongen die verliefd is op Marloes. Marloes weet dit, maar heeft absoluut geen gevoelens voor hem. Toch vindt Marloes de aandacht wel leuk en ze flirt er lustig op los. Na een avond uit met collega's zijn Marloes en de jongen alleen overgebleven. De jongen vraagt Marloes ineens blozend om met hem mee naar huis te gaan.
Marloes denkt even na en gaat lekker met hem mee naar huis en besluit hem de volgende ochtend te dumpen.
Later die week gaat Marloes naar de boekhandel. Het nieuwste deel van Marloes haar favoriete boekenserie is namelijk uit. Bovendien heeft de boekhandel vandaag een actie: de eerste 100 klanten krijgen het boek gratis. De winkel staat op het punt open te gaan en er staat al een grote menigte voor de deur. Marloes sluit achter aan en schat haar kansen in.
… Marloes is dolblij
or
verbolgen wanneer ze na een uur wachten
de 100ste klant is en het boek gratis krijgt
or
de 101ste klant is en de volle prijs moet betalen
62
Tessa loopt na de film naar de tramhalte twee straten verderop. Onderweg komt ze langs de Febo en ze beseft ineens dat ze honger heeft. Ze trekt een kroket uit de muur en loopt verder. Aan het einde van de straat slaat ze rechtsaf naar het plein. Ze steekt het
Tessa gooit achteloos het laatste stukje kroket naar het stel, raakt de ene vol en roept 'vieze homo's'.
Een korte tramrit later is Tessa bijna thuis. Ze neemt de lift naar haar appartement op de vijfde etage. In de lift ziet ze dat één van haar veters los is en ze bukt om die te strikken. Als ze de lift uitkomt en de galerij op loopt ziet ze haar voordeur open staan.
… Tessa is laaiend
or
uitgelaten als ze naar binnen rent en ziet dat
alles overhoop ligt en dure dingen zijn gejat
or
er niks weg is, wat een geluk!
164
plein schuin over naar de halte. Op het plein passeert ze twee mannen die hand in hand lopen en heel gelukkig lijken.
Ze herinnert zich in een flits dat ze de deur zelf niet goed dicht heeft gedaan.
63
Jasper fietst 's avonds na een etentje bij vrienden naar huis. Op een gegeven moment nadert Jasper een kruispunt. Van links komt een meisje ook op het kruispunt af, ze is zichtbaar gehaast. Jasper heeft voorrang en steekt dus het kruispunt over. Het meisje ziet Jasper niet aankomen en rijdt ook door. Ze schrikt wanneer ze pas op het laatste moment Jasper ziet en valt met haar fiets op straat.
Jasper fietst door en roept in het voorbijgaan dat ze dan maar de verkeersregels had moeten leren, domme trut.
Jasper fietst verder en ziet in de verte een kruispunt met verkeerslichten. Hij stopt wanneer hij er aankomt, omdat het licht nog steeds op rood staat. Het licht blijft maar op rood staan, en Jasper wordt ongeduldig. Er komt niks aan en het kruispunt lijkt verlaten. Hij rijdt door het rode licht, maar ziet plots aan de overkant een politieagent staan.
… Jasper is opgelucht
or
chagrijnig wanneer hij op de agent affietst
en deze knipoogt en zegt dat hij door mag rijden
or
en deze hem toch een fikse boete geeft
64
Claudia heeft boodschappen gedaan en loopt naar de auto. Ze laadt haar boodschappen in en stapt in. Claudia start de auto, zet de radio aan en geeft gas. Claudia heeft echter niet door dat de auto in zijn achteruit staat. De auto schiet naar achteren en knalt tegen de auto die
Claudia kijkt om zich heen, ziet dat gelukkig niemand het incident gezien heeft, stapt in en rijdt snel weg.
Die avond staat Claudia bij bioscoop Tuschinski in Amsterdam. Er is een grote internationale filmpremière met veel beroemde acteurs. Claudia hoopt bij de rode loper een glimp van de sterren op te vangen. Tot de aanwezige sterren
… Claudia is uitzinnig
or
teleurgesteld
als Johnny Depp vervolgens
rustig de tijd neemt om met haar te kletsen
or
haar nors negeert en ze zelfs geen handtekening krijgt
165
erachter geparkeerd staat. Claudia stapt uit en ziet dat de auto achter haar erg beschadigd is, maar haar eigen auto niet
behoort ook Johnny Depp, een acteur die Claudia erg bewondert. Op een gegeven moment ziet Claudia hem op de rode loper verschijnen.
A2: Study 2 & 3
Character Manipulation
Segment 4
Segment 5
Affective State Adjective
Neutral Segment Affect Reason
1
Sandra is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Sandra is dolblij
or
woedend als ze de gegevens doorleest
en haar aandelen in waarde verdubbbeld zijn
or
en haar aandelen niks meer waard blijken te zijn
2
Wesley is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Wesley is blij
or
geërgerd wanneer hij de post doorbladert
en de dvd erbij zit en hij dus door kan kijken or
de dvd er niet bijzit, maar wel een boete
3
Jeffrey is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Jeffrey voelt opluchting
or
frustratie wanneer hij bij de auto aan komt lopen
en deze geen beschadigingen blijkt te hebben
or
en die onder de kleine deuken zit van de hagelstenen
4
Emma is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Emma voelt blijdschap
or
woede opkomen wanneer het groepje
haar gelijk welkom doet voelen
or
haar aanstaart en vervolgens negeert
5
Ilse is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Ilse is dankbaar
or
pissig wanneer de medewerker haar ziet
en toch nog even snel binnenlaat
or
haar vuil aankijkt en de deur snel sluit
6
Martijn is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Martijn is euforisch
or
kwaad wanneer hij op het perron komt en
op tijd blijkt te zijn en de juiste trein haalt or
hoort omroepen dat er geen treinen rijden
7
Maria is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed
Maria is nijdig
or
opgelucht als de klodder gemorste saus
op haar nieuwe blouse valt en dus verpest is
or
op tafel valt en niet op haar nieuwe
166
mens OR echt een slecht mens
blouse
8
Stefan is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Stefan is boos
or
opgetogen
wanneer in het gedrang voor de inagng
iemand hem zonder pardon omver duwt or
iemand hem vriendelijk voor laat gaan
9
Sanne is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Sanne is verontwaardigd
or
opgetogen
wanneer de taxi vlak voor haar stopt
en iemand haar opzij duwt en gauw instapt or
en ze in kan stappen voor ze doorweekt is
10
Lieke is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Lieke constateert
nors
or
opgetogen
dat de salarisadminstratie
weer eens heeft gefaald, nog geen geld or
heeft gezorgd dat haar salaris op tijd is dit keer
11
Sarah is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Sarah constateert
pissig
or
opgelucht dat de laptop als ze hem weer aanzet
helemaal niks meer doet, compleet kapot
or
het weer gewoon blijkt te doen, gelukkig!
12
Danny is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Danny voelt frustratie
or
opluchting
dat het binnengekomen onderdeel
weer het verkeerde blijkt, dit kost hem geld
or
nu wel het goede blijkt te zijn, eindelijk!
13
Kirsten is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Kirsten voelt blijdschap
or
boosheid wanneer in de winkel blijkt dat
de tv zelfs in de aanbieding is en ze geld overhoudt
or
de tv die ze wil niet meer op voorraad blijkt te zijn
14
Judith is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Judith voelt woede
or
opluchting
als ze zich omdraait en
ziet hoe een man wegrent met haar portemonnee
or
een man haar portemonnee teruggeeft
15
Adriana is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Adriana voelt blijdschap
or
boosheid als er even later iemand langskomt
die haar vriendelijk naar de auto helpt or
en diegene haar vraag om hulp compleet negeert
16
Kelly is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Kelly is furieus
or
dolblij als ze in de email leest dat ze
niet mag herkansen en niet kan afstuderen or
een herkansing mag doen en kan afstuderen
17
Kim is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een
Kim is opgewekt
or
gefrustreerd
wanneer blijkt dat haar aanslag
nu eindelijk wel klopt
or
nog steeds niet klopt
167
slecht mens
18
David is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
David is chagrijnig
or
opgelucht als hij ziet hoe het apparaat
van de steiger valt en op de tegels kapot gaat
or
niet van de steiger valt en nog heel is
19
Michael is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Michael is verheugd
or
furieus wanneer bij aankomst blijkt dat het
nog mooier is dan gedacht en het weer schitterend is
or
appartement niet bestaat en hij nergens heen kan
20
Wendy is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Wendy is kwaad
or
vrolijk wanneer ze even later
merkt dat ze haar belachelijk maakten or
ook bij de grap wordt betrokken en meelacht
21
Laura is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Laura is trots
or
woedend wanneer blijkt dat haar aanvraag
is goedgekeurd en haar project kan starten
or
is afgekeurd om een enorm domme reden
22
Leonie is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Leonie voelt blijdschap
or
irritatie wanneer het garderobepersoneel
haar spullen snel vindt en ze droog thuiskomt
or
pas na tien minuten haar spullen vindt en ze zeiknat wordt
23
Thomas is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Thomas is furieus
or
verheugd als de man omkijkt, even wacht en
dan de deur in zijn gezicht dicht laat vallen or
de deur vriendelijk glimlachend open houdt
24
Nina is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Nina is opgelucht
or
verbolgen als ze aankomt op het vliegveld en
blijkt dat haar vlucht ook vertraagd is or
blijkt dat ze te laat is om nog te boarden
25
Koen is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Koen is uitzinnnig
or
laaiend wanneer de jongen op hem afloopt
en zijn sleutels gevonden blijkt te hebben
or
ineens Koens tas grijpt en hard wegrent
26
Cynthia is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Cynthia is kwaad
or
opgelucht als uiteindelijk blijkt dat
de resultaten zoek zijn en ze opnieuw moet
or
de resultaten goed zijn, ze is kerngezond
27
Frank is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed
Frank is ziedend
or
blij als de ober reageert op een
boze manier en net doet of hij zeurt or
vriendelijke en correcte manier
168
mens OR echt een slecht mens
28
Remco is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Remco is gefrustreerd
or
verheugd als hij na een paar minuten rijden
geen bezine meer heeft en stil komt te staan or
een bezinestation ziet en hij kan tanken
29
Maarten is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Maarten voelt razernij
or
blijdschap als hij eenmaal thuiskomt
en zijn vrouw met ander in bed aantreft
or
en ze samen een romantische ochtend hebben
30
Jordy is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Jordy is pissig
or
blij als de buschauffeur hem ziet
en gewoon doorrijdt en hem daar laat staan or
en vriendelijk even wacht zodat hij mee kan
31
Ruben is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Ruben wordt witheet
or
vrolijk wanneer de man voor hem
hem een duw geeft en hij dus keihard valt or
hem vriendelijk helpt overeind te blijven
32
Dennis is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Dennis is kwaad
or
dankbaar als hij de tuin inkijkt en ziet dat
de buren hun tuinafval in zijn tuin hebben gegooid
or
de buurman al het werk voor hem heeft gedaan
33
Saskia is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Saskia is laaiend
or
uitzinnig wanneer de baas in het gesprek
haar onterecht dingen aanrekent en ontslaat
or
vertelt dat ze promotie krijgt en opslag
34
Tim is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Tim is blij
or
geïrriteerd
wanneer de cake uiteindelijk
perfect uit de oven komt
or
aangebrand uit de oven komt
35
Joris is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Joris is woest
or
blij als hij een mailtje krijgt met daarin
informatie die hij 3 maanden geleden nodig had
or
eindelijk het verlossende woord
36
Mark is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Mark is trots
or
boos als hij aanlegt om te schieten
en de keeper passeert en scoort
or
en hij keihard onderuit wordt gehaald
37
Myrthe is een type P persoonlijkheid OR type O persoonlijkheid OR
Myrthe is opgelucht
or
getergd als ze bij de buren aanbelt en
die thuis blijken te zijn in ze dus haar huis in
or
die op vakantie zijn en ze een
169
echt een goed mens OR echt een slecht mens
kan slotenmaker moet bellen
38
Bram is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Bram is gelukkig
or
razend als hij aankomt waar hij zijn fiets had staan
en ziet dat deze er gewoon nog staat or
en hij echt gejat is en hij naar huis kan lopen
39
Kevin is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Kevin is uitgelaten
or
pissig wanneer de andere automobilist
stopt en zijn auto kan repareren
or
wel langzamer gaat rijden, maar niet stopt
40
Anouk is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Anouk is driftig
or
opgetogen
als uit de brief blijkt dat ze
wederom een groot bedrag moet bijbetalen or
dit jaar zelf een fiks bedrag terugkrijgt
41
Jesse is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Jesse is woest
or
uitgelaten wanneer vlak voor het einde
de tegenpartij scoort en hij zijn geld kwijt is
or
zijn team scoort en hij dus 300 euro wint
42
Inge is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Inge voelt vreugde
or
woede wanneer ze te horen krijgt
dat haar bod op een huis is geaccepteerd
or
dat er bij haar is ingebroken en alles weg is
43
Julia is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Julia is geïrriteerd
or
vrolijk als de andere klant het spelletje
snel afrekent en haar compleet negeert or
aan haar geeft zodat ze haar cadeau heeft
44
Larissa is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Larissa is gerustgesteld
or
laaiend als ze twee uur later naar buiten loopt
met een nieuwe en zeer gunstige hypotheek
or
zonder resultaat omdat ze haar afspraak waren vergeten
45
Hendrik is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Hendrik is dankbaar
or
getergd als de auto's die nu groen hebben
rustig wachten en hem vriendelijk de tijd geven
or
ongeduldig beginnen te toeteren en optrekken
46
Eva is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Eva is dankbaar
or
woedend als de conducteur haar ziet rennen
en vriendelijk even op haar wacht or
de deuren expres voor haar neus sluit
47
Wouter is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed
Wouter is tevreden
or
woedend als bij het monteren blijkt
dat de kast perfect past waar hij hem wilde
or
dat de planken scheef zijn en er dingen
170
mens OR echt een slecht mens
ontbreken
48
Elise is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Elise wordt chagrijnig
or
vrolijk als de serveerster reageert
op een norse manier en haar boos aankijkt or
op een vriendelijke manier en ze kletsen wat
49
Tom is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Tom is verrast
or
pissig wanneer de telefoon over gaat
en hij gelijk de assistente krijgt
or
en hij de 15de wachtende blijkt te zijn
50
Niels is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Niels voelt razernij
or
blijdschap wanneer zijn manager in de meeting
liegt en alle eer voorzichzelf opstrijkt or
hem prijst voor zijn originaliteit en inzet
51
Mike is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Mike is dankbaar
or
gepikeerd wanneer de man die ook wil rennen
hem vriendelijk voor laat gaan or
net voor hem op de band springt
52
Rianne is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Rianne wordt woest
or
vrolijk als ze leest dat iedereen
zegt niet te kunnen terwijl ze weet dat ze wel uitgaan
or
enthousiast is en ze een leuke avond heeft
53
Bart is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Bart is opgelucht
or
furieus wanneer zijn manager hem vertelt
dat hij een promotie en opslag krijgt
or
dat hij per direct ontslag krijgt
54
Sander is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Sander wordt boos
or
blij als het groepje jongeren
expres de weg blijft blokkeren en lacht or
gauw aan de kant gaat zodat hij door kan lopen
55
Tamara is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Tamara wordt furieus
or
blij als de klant die haar aanstootte
geen sorry zegt en zelfs geamuseerd omkijkt or
aardig blijkt te zijn en helpt op te ruimen
56
Thijs is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Thijs is ziedend
or
blij als hij ziet hoe de omstanders
domweg opzij gaan en de dief ontsnapt or
de dief grijpen voordat deze weg kan komen
57
Jeroen is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een
Jeroen is witheet
or
vergenoegd
als de ober vervolgens
zonder wat te zeggen omdraait en hem negeert
or
geduldig zijn best doet om Engels te spreken
171
slecht mens
58
Patrick is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Patrick is opgelucht
or
ziedend wanneer na het opnieuw opstarten blijkt
dat er een automatische back‐up beschikbaar is
or
dat het document verloren is en het opnieuw moet
59
Anna is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Anna wordt vrolijk
or
kwaad als de parkeerwachter haar ziet
en besluit de boete te verscheuren
or
en toch alsnog de boete uitschrijft
60
Nienke is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Nienke is gepikeerd
or
gelukkig als de man vervolgens zegt dat
ze zonder bonnetje naar haar geld kan fluiten
or
ze ook zonder bonnetje haar geld terugkrijgt
61
Marloes is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Marloes is dolblij
or
verbolgen wanneer ze na een uur wachten
de 100ste klant is en het boek gratis krijgt
or
de 101ste klant is en de volle prijs moet betalen
62
Tessa is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Tessa is laaiend
or
uitgelaten als ze naar binnen rent en ziet dat
alles overhoop ligt en dure dingen zijn gejat
or
er niks weg is, wat een geluk!
63
Jasper is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Jasper is opgelucht
or
chagrijnig wanneer hij op de agent affietst
en deze knipoogt en zegt dat hij door mag rijden
or
en deze hem toch een fikse boete geeft
64
Claudia is een type P persoonlijkheid OR type O persoonlijkheid OR echt een goed mens OR echt een slecht mens
Claudia is uitzinnig
or
teleurgesteld
als Johnny Depp vervolgens
rustig de tijd neemt om met haar te kletsen
or
haar nors negeert en ze zelfs geen handtekening krijgt
172
Supplementary Information B
B1: Study 1
173
B2: Study 2
174
B3 ‐ Morality Manipulation: Study 3
175
B3 – Minimal Group Manipulation: Study 3
176
Supplementary Information C
C1.1
Model summary Character Manipulation: Study 1
Character Morality Corrugator
total cases cases after baseline rejection
data loss
191900 190950 0.50%
Iterative model report for Character Morality Manipulation. Each line reports the assessment of improved model fit after adding a single predictor.
Nr. ‐2 LL nr of parameters
p model fit (chisquare distribution)
model compa‐rision
predictor added action
Model 0 2297839.064 2 empty model ‐
Model 1 2288296.555 3 0.000 better Subject Random Intercept
keep
Model 2 2279588.985 4 0.000 better Item Random Intercept
keep
Model 3 2279517.602 5 0.000 better Moral Linear Time Fixed
keep
Model 4 2277709.547 6 0.000 better Immoral Linear Time Fixed
keep
Model 5 2277553.13 8 0.000 better Moral Linear Time Random Slope (Subject UN)
keep
Model 6 2272924.031 11 0.000 better Immoral Linear Time Random Slope (Subject UN)
keep
Model 7 2272922.516 12 0.218 not better
Moral Quadratic Time Fixed
remove
Model 8 2272791.796 12 0.000 better Immoral Quadratic Time Fixed
keep
Model 9 2272789.978 13 0.178 not better
Moral Cubic Fixed remove
Model 10 2272633.675 13 0.000 better Immoral Cubic Time Fixed
keep
Model 11 2272502.391 14 0.000 better Character Morality Conditie
keep
Model 12 2270884.847 18 0.000 better Immoral Quadratic Time Random Slope (Subject UN)
keep
Model 13 2270262.398 23 0.000 better Immoral Cubic Time Random Slope (Subject UN)
keep
Model 14 2270261.652 24 0.388 not better
Partcipant Gender remove
177
C1.2
Parameter Estimates Character Manipulation: Study 1
Estimates of Fixed Effects Character Moralitya Parameter Estimate Std. Error df t Sig. 95% Confidence Interval
Lower Bound
Upper Bound
Linear Moral ‐1.795 0.446 59.986 ‐4.021 0.000 ‐2.688 ‐0.902 Linear Immoral 15.140 3.330 59.969 4.547 0.000 8.480 21.800 Quadratic Immoral ‐2.020 0.698 62.428 ‐2.895 0.005 ‐3.415 ‐0.626 Cubic Immoral ‐1.623 0.448 59.877 ‐3.620 0.001 ‐2.520 ‐0.726 Immoral 138.173 3.007 108.982 45.949 0.000 132.213 144.133 Moral 108.717 2.989 106.320 36.378 0.000 102.792 114.642 a. Dependent Variable: Character Morality Corrugator Response.
Pairwise Comparisons Main Effect of Character Moralitya
(I) likeability manipulation
Mean Difference (I‐J) Std. Error df Sig.c
95% Confidence Interval for Differencec Lower Bound
Upper Bound
moral immoral ‐29,456* 2.262 264.421 0.000 ‐33.909 ‐25.003 Based on estimated marginal means *. The mean difference is significant at the ,05 level. a. Dependent Variable: Character Morality Corrugator Response. c. Adjustment for multiple comparisons: Bonferroni. Pairwise Comparison between Linear Estimates Moral & Immoral
a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval
Lower Bound
Upper Bound
Linear Moral vs. Linear Immoral ‐16.928 3.230 62.122 0.000 ‐5.241 0.000 ‐23.385 ‐10.472 a. linear moral vs immoral b. Dependent Variable: Character Morality Corrugator Response.
178
C1.3
Model Summary Critical Event: Study 1
Critical Event Corrugator
total cases cases after baseline rejection
data loss
191900.000 190950.000 0.005
Iterative model report for Critical Event Manipulation. Each line reports the assessment of improved model fit after adding a single predictor.
Nr. ‐2 LL nr of parameters p model fit (chisquare distribution)
model comparision
predictor added action
Model 0 2194785.534 2.000 empty model
Model 1 2183526.383 3.000 0.000 better Subject Random Intercept
keep
Model 2 2176247.842 4.000 0.000 better Item Random Intercept
keep
Model 3 2176121.731 5.000 0.000 better Linear Time Moral Positive Fixed
keep
Model 4 2176116.686 6.000 0.025 better Linear Time Immoral Positive Fixed
keep
Model 5 2176103.552 7.000 0.000 better Linear Time Immoral Negative Fixed
keep
Model 6 2176011.053 8.000 0.000 better Linear Time Moral Negative Fixed
keep
Model 7 2175796.515 10.000 0.000 better Linear Time Moral Positive Random Slope (Subject UN)
keep
Model 8 2175386.242 13.000 0.000 better Linear Time Immoral Positive Random Slope (Subject UN)
keep
Model 9 2174903.380 17.000 0.000 better Linear Time Immoral Negative Random Slope (Subject UN)
keep
Model 10 2173945.400 22.000 0.000 better Linear Time Moral Negative Random Slope (Subject UN)
keep
Model 11 2173915.017 23.000 0.000 better Quadratic Time Moral Positive Fixed
keep
Model 12 2173903.890 24.000 0.001 better Quadratic Time Immoral Positive Fixed
keep
Model 13 2173899.692 25.000 0.040 better Quadratic Time Immoral Negative Fixed
keep
Model 14 2173855.657 26.000 0.000 better Quadratic Time Moral Negative Fixed
keep
Model 15 2172900.165 32.000 0.000 better Quadratic Time keep
179
Moral Positive Random (Subject UN)
Model 16 2172408.175 39.000 0.000 better Quadratic Time Immoral Positive Random (Subject UN)
keep
Model 17 2171607.151 47.000 0.000 better Quadratic Time Immoral Negative Random (Subject UN)
keep
Model 18 2171061.953 56.000 0.000 better Quadratic Moral Negative Random (Subject UN)
keep
Model 19 2171061.828 57.000 0.724 not better Cubic Time Moral Positive Fixed
remove
Model 20 2171061.634 57.000 0.572 not better Cubic Time Immoral Positive Fixed
remove
Model 21 2171061.656 57.000 0.586 not better Cubic Time Immoral Negative Fixed
remove
Model 22 2171054.394 57.000 0.006 better Cubic Time Moral Negative Fixed
keep
Model 23 0.000 0.000 0.000 no convergence Cubic Time Moral Negative Random (Subject UN)
remove
Model 24 2171054.349 58.000 0.832 not better Character Morality
keep for interaction
Model 25 2171029.814 59.000 0.000 better Event Valence keep
Model 26 2170994.271 60.000 0.000 better Character Morality * Event Valence
keep
Model 27 2170993.277 61.000 0.319 not better Participant Gender
remove
C1.4
Parameter Estimates Critical Event: Study 1
Estimates of Fixed Effects Critical Eventa Parameter Estimate Std. Error df t Sig. 95% Confidence Interval
Lower Bound
Upper Bound
Linear Moral Pos ‐2.578 0.568 59.742 ‐4.542 0.000 ‐3.713 ‐1.442 Linear Immoral Pos ‐0.491 0.714 59.897 ‐0.687 0.495 ‐1.920 0.938 Linear Immoral Neg ‐0.850 0.764 59.999 ‐1.112 0.271 ‐2.379 0.679 Linear Moral Neg 3.603 1.126 95.842 3.199 0.002 1.367 5.839 Quadratic Moral Pos 1.109 0.568 64.893 1.954 0.055 ‐0.024 2.242 Quadratic Immoral Pos 0.555 0.439 68.251 1.263 0.211 ‐0.322 1.431 Quadratic Immoral Neg 0.335 0.524 65.818 0.638 0.525 ‐0.712 1.381 Quadratic Moral Neg ‐1.319 0.484 66.479 ‐2.723 0.008 ‐2.285 ‐0.352 Cubic Moral Neg ‐0.378 0.137 190155.593 ‐2.749 0.006 ‐0.647 ‐0.108 Immoral Neg 115.331 3.195 122.173 36.102 0.000 109.007 121.655 Immoral Pos 116.696 3.195 122.174 36.530 0.000 110.372 123.020 Moral Neg 126.133 3.194 122.148 39.486 0.000 119.809 132.456 Moral Pos 105.027 3.195 122.180 32.876 0.000 98.703 111.351
180
a. Dependent Variable: Critical Event Corrugator Response.
Pairwise Comparisons Character Morality*EventValence
a
(I) reference like‐pos
Mean Difference (I‐J)
Std. Error df Sig.
c 95% Confidence Interval for Differencec Lower Bound
Upper Bound
immoralneg immoralpos ‐1.365 2.578 272.395 1.000 ‐8.217 5.487 moralneg ‐10.802 2.578 272.311 0.000 ‐17.653 ‐3.950 moralpos 10.304 2.578 272.414 0.000 3.452 17.156
immoralpos moralneg ‐9.437 2.578 272.313 0.002 ‐16.289 ‐2.585 moralpos 11.669 2.578 272.418 0.000 4.817 18.521
moralneg moralpos 21.106 2.578 272.333 0.000 14.254 27.958 Based on estimated marginal means *. The mean difference is significant at the ,05 level. a. Dependent Variable: Critical Event Corrugator Response. c. Adjustment for multiple comparisons: Bonferroni.
Pairwise Comparison between Linear Moral Pos & Linear Moral Nega,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval Lower Bound
Upper Bound
Linear Moral Pos vs. Linear Moral Neg ‐6.182 1.262 136.663 0.000 ‐4.899 0.000 ‐8.677 ‐3.687
C2.1
Model Summary Character Manipulation: Study 2
Character Morality Manipulation Corrugator
total cases cases after baseline
data loss
192000.000 190050.000 0.010
Nr. ‐2 LL nr of parameters p model fit (chi‐square distribution
model comparison
predictor added
action
Model 0 2529242.959 2.000 empty model
Model 1 2511977.094 3.000 0.000 better Subject Random
keep
Model 2 2501228.524 4.000 0.000 better Item Random
keep
Model 3 2501225.912 5.000 0.106 not better Moral Linear Fixed
remove
Model 4 2498686.196 5.000 0.000 better Immoral Linear Fixed
keep
Model 5 2491340.511 7.000 0.000 better Immoral Linear Random (Subject Unstructured)
keep
Model 6 2491339.829 8.000 0.409 not better Moral Quad Fixed
remove
181
Model 7 2490601.343 8.000 0.000 better Immoral Quad Fixed
keep
Model 8 2488946.375 11.000 0.000 better Immoral Quad Random (Subject Unstructured)
keep
Model 9 2488944.278 12.000 0.148 not better Moral Cube Fixed
remove
Model 10 2488818.493 12.000 0.000 better Immoral Cube Fixed
keep
Model 11 2488122.473 16.000 0.000 better Immoral Cube Random (Subject Unstructured)
keep
Model 12 2487937.942 17.000 0.000 better Character Morality
keep
C2.2
Parameter Estimates Character Manipulation: Study 2
Type III Tests of Fixed Effectsa
Source Numerator df Denominator df F Sig.
linear immoral 1.000 60.004 17.364 0.000
quadratic immoral 1.000 61.843 38.949 0.000
cubic immoral 1.000 60.030 9.505 0.003
Character Morality 2.000 111.913 369.144 0.000
a. Dependent Variable: Corrugator Response Character Morality
Estimates of Fixed Effectsa
Parameter Estimate Std. Error df t Sig. 95% Confidence Interval
Lower Bound
Upper Bound
Intercept (moral) 114.061 7.288 86.138 15.651 0.000 99.574 128.548
linear immoral 29.549 7.091 60.004 4.167 0.000 15.364 43.733
quadratic immoral ‐8.306 1.331 61.843 ‐6.241 0.000 ‐10.966 ‐5.645
cubic immoral ‐2.628 0.852 60.030 ‐3.083 0.003 ‐4.333 ‐0.923
immoral 74.022 4.513 261.865 16.402 0.000 65.135 82.908
moral 0b 0.000
a. Dependent Variable: Corrugator Response Character Morality.
b. This parameter is set to zero because it is redundant.
Iterative model report for Affective State Adjective ‐ Corrugator. Each line reports the assessment of model fit after adding a single predictor
182
C2.3
Model Summary Affective State Adjective: Study 2
Affective State Adjective Corrugator total cases cases after baseline data loss 38400.000 38010.000 0.010
Nr. ‐2 LL nr of parameters
p model fit (chi‐square distribution
model comparison predictor added action
Model 0 429997.771 2.000 empty model
Model 1 427395.692 3.000 0.000 better Subject Random keep
Model 2 425245.687 4.000 0.000 better Item Random keep
Model 3 425219.406 5.000 0.000 better Linear moral‐positive
keep
Model 4 425217.467 6.000 0.164 not better Linear immoral‐positive
remove
Model 5 425209.653 6.000 0.002 better Linear immoral‐negative
keep
Model 6 425183.699 7.000 0.000 better Linear moral‐negative
keep
Model 7 0.000 0.000 0.000 no convergence Linear moral‐positive random (Subject Unstructured)
remove
Model 8 425167.467 9.000 0.000 better Linear immoral negative random (Subject Unstructured)
keep
Model 9 424963.589 12.000 0.000 better Linear moral‐negative random (Subject Unstructured)
keep
Model 10 424963.420 13.000 0.681 not better Quadratic moral‐positive
remove
Model 11 424963.462 13.000 0.722 not better Quadratic immoral‐positive
remove
Model 12 424963.516 13.000 0.787 not better Quadratic immoral‐negative
remove
Model 13 424963.582 13.000 0.933 not better Quadratic moral‐negative
remove
Model 14 424960.124 13.000 0.063 not better Cubic moral‐positive
remove
Model 15 424962.323 13.000 0.261 not better Cubic immoral‐positive
remove
Model 16 424963.567 13.000 0.882 not better Cubic immoral‐negative
remove
Model 17 424959.308 13.000 0.039 better Cubic moral‐negative
keep
Model 18 0.000 0.000 0.000 no convergence Cubic moral‐negative random (Subject Unstructured)
remove
Model 19 424957.396 14.000 0.167 not better Character Morality keep for interaction
Model 19a 424948.906 15.000 0.004 beter Affective State Adjective Valence
keep
Model 19b 424947.957 16.000 0.330 niet beter Character Morality * Affective State Adjective Valence
remove
Model 20 424950.883 14.000 0.004 beter Valence keep
183
C2.4
Parameter Estimates Affective State Adjective: Study 2
Type III Tests of Fixed Effectsa Source Numerator df Denominator df F Sig. linear moral‐positive 1.000 37578.694 26.543 0.000 linear moral‐negative 1.000 244.524 9.146 0.003 linear immoral‐negative 1.000 60.641 5.082 0.028 cubic moral‐negative 1.000 37578.694 4.281 0.039 Affective State Adjective Valence 2.000 125.800 1080.456 0.000 a. Dependent Variable: Corrugator Response Affective State Adjective
Estimates of Fixed Effectsa Parameter Estimate Std. Error df t Sig. 95% Confidence Interval Lower
Bound Upper Bound
Intercept 118.742 2.869 115.118 41.382 0.000 113.059 124.426 linear moral‐positive ‐11.746 2.280 37578.694 ‐5.152 0.000 ‐16.215 ‐7.277 linear moral‐negative 23.247 7.687 244.524 3.024 0.003 8.106 38.389 linear immoral‐negative ‐7.176 3.183 60.641 ‐2.254 0.028 ‐13.542 ‐0.810 cubic moral‐negative ‐77.340 37.378 37578.694 ‐2.069 0.039 ‐150.603 ‐4.078 Affective State Adjective Negative
6.697 2.288 252.702 2.927 0.004 2.191 11.204
Affective State Adjective Positive
0b 0.000
a. Dependent Variable: Corrugator Response Affective State Adjective
b. This parameter is set to zero because it is redundant.
Custom Hypothesis a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 ‐34.993 8.018 289.254 0.000 ‐4.364 0.000 ‐50.775 ‐19.212 a. linear moral‐positive vs. linear moral‐neg
b. Dependent Variable: Corrugator Response Affective State Adjective
Custom Hypothesis a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 ‐4.570 3.915 138.756 0.000 ‐1.167 0.245 ‐12.312 3.171 a. linear moral‐positive vs. linear immoral‐negative
b. Dependent Variable: Corrugator Response Affective State Adjective
Custom Hypothesis a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 30.423 7.539 266.338 0.000 4.035 0.000 15.580 45.267 a. linear moral‐negative vs. linear immoral‐negative
b. Dependent Variable: Corrugator Response Affective State Adjective
184
C2.5
Model Summary Affect Reason: Study 2
Affect Reason Corrugator
total cases cases after baseline data loss
96000.000 95025.000 0.010
Nr. ‐2 LL nr of parameters
p model fit (chi‐square distribution
model comparison predictor added
action
Model 0 1180252.740 2.000 empty model
Model 1 1175596.403 3.000 0.000 better Subject Random
keep
Model 2 1170435.975 4.000 0.000 better Item Random keep
Model 3 1170424.193 5.000 0.001 better Linear moral‐positive
keep
Model 4 1170417.344 6.000 0.009 better Linear immoral‐positive
keep
Model 5 1170416.625 7.000 0.396 not better Linear immoral‐negative
remove
Model 6 1170273.289 7.000 0.000 better Linear moral‐negative
keep
Model 7 1170272.627 9.000 0.718 not better Linear moral‐positive random (Subject Unstructured)
remove
Model 8 0.000 0.000 0.000 no convergence Linear immoral‐positive random (Subject Unstructured)
remove
Model 9 1169292.229 9.000 0.000 better Linear moral‐negative random (Subject Unstructured)
keep
Model 10 1169292.000 10.000 0.632 not better Quadratic moral‐positive
remove
Model 11 1169292.164 10.000 0.799 not better Quadratic immoral‐positive
remove
Model 12 1169292.085 10.000 0.704 not better Quadratic immoral‐negative
remove
Model 13 1169275.080 10.000 0.000 not better Quadratic moral‐negative
keep
Model 14 1168366.221 13.000 0.000 better Quadratic moral‐negative random (Subject Unstructured)
keep
Model 15 1168365.266 14.000 0.328 not better Cubic moral‐positive
remove
Model 16 1168366.217 14.000 0.950 not better Cubic remove
185
immoral‐positive
Model 17 1168363.831 14.000 0.122 not better Cubic immoral‐negative
remove
Model 18 1168358.216 14.000 0.005 better Cubic moral‐negative
keep
Model 19 1168306.203 18.000 0.000 better Cubic moral‐negative random (Subject Unstructured)
keep
Model 20 1168302.547 19.000 0.056 marginally better Character Morality
keep for interaction
Model 21 1168257.624 20.000 0.000 better Affect Reason Valence
keep
Model 22 1168233.039 21.000 0.000 better Character Morality * Affect Reason Valence
keep
C2.6
Parameter Estimates Affect Reason: Study 2
Type III Tests of Fixed Effectsa Source Numerator df Denominator df F Sig. linear moral‐positive 1.000 94530.935 12.092 0.001 linear immoral‐positive 1.000 94530.935 7.028 0.008 linear moral‐negative 1.000 60.001 8.176 0.006 quadratic moral‐negative 1.000 66.281 3.025 0.087 cubic moral‐negative 1.000 60.073 2.831 0.098 Character Morality * Affect Reason Valence
4.000 178.902 362.775 0.000
a. Dependent Variable: Corrugator Response Affect Reason
Estimates of Fixed Effectsa Parameter Estimate Std. Error df t Sig. 95% Confidence Interval Lower
Bound Upper Bound
linear moral‐positive ‐3.506 1.008 94530.935 ‐3.477 0.001 ‐5.481 ‐1.530 linear immoral‐positive ‐2.673 1.008 94530.935 ‐2.651 0.008 ‐4.649 ‐0.697 linear moral‐negative 18.907 6.612 60.001 2.859 0.006 5.681 32.133 quadratic moral‐negative ‐8.590 4.939 66.281 ‐1.739 0.087 ‐18.449 1.270 cubic moral‐negative ‐7.095 4.217 60.073 ‐1.682 0.098 ‐15.530 1.340 immoral‐negative 129.905 4.528 205.263 28.688 0.000 120.978 138.833 moral‐negative 154.699 4.601 218.813 33.621 0.000 145.631 163.767 immoral‐positive 122.120 4.528 205.150 26.972 0.000 113.193 131.046 moral‐positive 112.942 4.528 205.150 24.945 0.000 104.015 121.868 a. Dependent Variable: Corrugator Response Affect Reason
Custom Hypothesis a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 ‐22.413 6.689 62.823 0.000 ‐3.351 0.001 ‐35.779 ‐9.046 a. linear moral‐positive vs linear moral‐negative‐neg
186
b. Dependent Variable: Corrugator Response Affect Reason
Custom Hypothesis
a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 ‐0.833 1.426 94530.935 0.000 ‐0.584 0.559 ‐3.627 1.961 a. linear moral‐positive vs linear immoral‐positive
b. Dependent Variable: Corrugator Response Affect Reason
Custom Hypothesis a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 ‐21.580 6.689 62.823 0.000 ‐3.226 0.002 ‐34.947 ‐8.213 a. linear moral‐negative vs linear immoral‐positive
b. Dependent Variable: Corrugator Response Affect Reason
Custom Hypothesis
a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 41.757 4.765 265.686 0.000 8.764 0.000 32.376 51.139 a. moral‐positive vs moral‐negative
b. Dependent Variable: Corrugator Response Affect Reason
Custom Hypothesis
a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 7.786 4.694 250.326 0.000 1.659 0.098 ‐1.459 17.031 a. immoral‐positive vs immoral‐negative
b. Dependent Variable: Corrugator Response Affect Reason
Custom Hypothesis
a,b
Contrast Estimate Std. Error df Test Value t Sig. 95% Confidence Interval Lower
Bound Upper Bound
L1 9.178 4.694 250.197 0.000 1.955 0.052 ‐0.066 18.422 a. moral‐positive vs immoral‐positive
b. Dependent Variable: Corrugator Response Affect Reason
Iterative model report for Neutral Segment ‐ Corrugator. Each line reports the assessment of model fit after adding a single predictor
187
C2.6
Neutral Segment: Study 2
Neutral Segment Corrugator
total cases cases after baseline
data loss
115200.000 114030.000 0.010
Nr. ‐2 LL nr of parameters
p model fit (chi‐square distribution
model comparison
predictor added
action
Model 0 1330985.759 2.000 leeg model
Model 1 1322707.229 3.000 0.000 better Subject Random
keep
Model 2 1315978.492 4.000 0.000 better Item Random keep
Model 3 1315978.459 5.000 0.856 not better Character Morality
keep
Model 4 1315965.427 6.000 0.000 better Valence keep
Model 5 1315950.014 7.000 0.000 better Character Morality * Valence
keep
Model 6 1310801.532 8.000 0.000 better Like Conditie * Event Valence Random (Subject VC)
keep
Type III Tests of Fixed Effectsa Source Numerator df Denominator df F Sig. like_by_valence 4.000 195.094 408.613 0.000 a. Dependent Variable: valence_response.
Estimates of Fixed Effectsa Parameter Estimate Std. Error df t Sig. 95% Confidence Interval Lower
Bound Upper Bound
immoral‐negative 124.154 4.378 255.531 28.358 0.000 115.532 132.775 moral‐negative 134.484 4.378 255.509 30.719 0.000 125.863 143.106 immoral‐positive 124.635 4.378 255.473 28.470 0.000 116.014 133.256 moral‐positive 115.312 4.378 255.473 26.340 0.000 106.691 123.933 a. Dependent Variable: Corrugator Response Neutral Segment.
Custom Hypothesis Test 1 (immoral‐pos vs. immoral‐neg)
Contrast Estimatesa,b Contrast Estimate Std.
Error df Test Value t Sig. 95% Confidence Interval
Lower Bound
Upper Bound
L1 ‐0.481 5.053 382.366 0.000 ‐0.095 0.924 ‐10.417 9.454 a. immoral‐pos vs. immoral‐neg
b. Dependent Variable: valence_response
Custom Hypothesis Test 2 (moral‐pos vs. moral‐neg)
Contrast Estimatesa,b Contrast Estimate Std.
Error df Test Value t Sig. 95% Confidence Interval
Lower Bound
Upper Bound
L1 19.173 5.053 382.340 0.000 3.794 0.000 9.237 29.108
188
a. moral‐pos vs. moral‐neg
b. Dependent Variable: valence_response
C3.1
Model Summary Character Manipulation: Study 3
total cases cases after baseline data loss Character Manipulation Corrugator 277760.000 268170.000 0.035
‐2 LL nr of parameters
p model fit (chi‐square distribution
model comparision
predictor added action
Model 0 2908526.089 2.000 Intercept Only
Model 1 2902559.555 3.000 0.000 better Subject Random keep
Model 2 2897409.341 4.000 0.000 better Item Random keep
Model 3 2897381.986 7.000 0.000 better Like Conditie Fixed keep
Model 4 2887639.735 16.000 0.000 better Like Conditie Random keep
Model 4 2887638.017 17.000 0.190 not better Moral Linear Fixed remove
Model 4 2887518.662 17.000 0.000 better Immoral Linear Fixed keep
Model 5 2887483.288 18.000 0.000 better Ingroup Linear Fixed keep
Model 6 2887460.261 19.000 0.000 better Outgroup Linear Fixed keep
Model 7 2885750.824 24.000 0.000 better Immoral Linear Random keep
Model 8 2885292.822 30.000 0.000 better Ingroup Linear Random keep
Model 9 2885113.681 37.000 0.000 better Outgroup Linear Random keep
Model 10 2885110.117 38.000 0.059 not better Moral Quadratic Fixed remove
Model 11 2884947.628 38.000 0.000 better Immoral Quadratic Fixed keep
Model 12 2884942.294 39.000 0.021 better Ingroup Quadratic Fixed remove
Model 13 2884942.177 40.000 0.732 not better Outgroup Quadratic Fixed keep
Model 14 2882090.422 47.000 0.000 better Immoral Quad Random keep
Model 14 0.000 0.000 0.000 no convergence Ingroup Quad Random remove
Model 15 2882088.651 48.000 0.183 not better Moral Cube Fixed remove
Model 16 2882087.577 48.000 0.092 not better Immoral Cube Fixed remove
Model 17 2882090.419 48.000 0.956 not better Ingroup Cube Fixed remove
Model 8 2882082.145 48.000 0.004 better Outgroup Cube Fixed keep
Model 19 0.000 0.000 0.000 no convergence Outgroup Cube Random remove
Model 20 2882081.506 49.000 0.424 not better Block Order remove
189
C3.2
Parameter Estimates Character Manipulation: Study 3
Estimates of Fixed Effectsa
Parameter Estimate Std. Error df t Sig.
95% Confidence Interval
Lower Bound Upper Bound
Moral 103.930 1.333 112.504 77.992 0.000 101.290 106.570
Immoral 111.961 4.513 64.933 24.806 0.000 102.947 120.975
Ingroup 104.287 1.279 130.894 81.525 0.000 101.756 106.817
Outgroup 104.397 1.117 152.774 93.425 0.000 102.189 106.604 Linear Immoral 2.191 1.111 62.017 1.972 0.053 ‐0.030 4.412 Linear Ingroup 1.192 0.627 61.900 1.901 0.062 ‐0.062 2.446 Linear Outgroup 2.260 0.622 299.092 3.634 0.000 1.036 3.484 Quadratic Immoral ‐2.883 1.540 61.983 ‐1.873 0.066 ‐5.961 0.195 Quadratic Ingroup 0.509 0.219 267175.317 2.322 0.020 0.079 0.939 Cubic Outgroup ‐0.718 0.250 267175.317 ‐2.877 0.004 ‐1.208 ‐0.229 a. Dependent Variable: Corrugator Response Character Manipulation
Estimatesa
likeConditie Mean Std. Error df
95% Confidence Interval
Lower Bound Upper Bound
Moral 103,328b 1.446 101.694 100.460 106.197
Immoral 111,359b 4.140 65.535 103.092 119.626
Ingroup 103,685b 1.343 116.767 101.026 106.344
Outgroup 103,795b 1.213 130.985 101.396 106.195
a. Dependent Variable: Corrugator Response Character Manipulation b. Covariates appearing in the model are evaluated at the following values: t1_immoral = ,0000, t1_ingroup = ,0000, t1_outgroup = ,0000, t2_immoral = ,2540, t2_ingroup = ,2569, t3_outgroup = ,0000.
Pairwise Comparisonsa
(I) likeConditie
Mean Difference (I‐J) Std. Error df Sig.b
95% Confidence Interval for Differenceb Lower Bound Upper Bound
Moral
Immoral ‐8.031 4.752 67.394 0.574 ‐20.949 4.887
Ingroup ‐0.357 1.400 224.467 1.000 ‐4.084 3.370
Outgroup ‐0.467 1.319 245.921 1.000 ‐3.975 3.041
Immoral
Moral 8.031 4.752 67.394 0.574 ‐4.887 20.949
Ingroup 7.674 4.634 68.080 0.614 ‐4.920 20.268
Outgroup 7.564 4.628 67.760 0.641 ‐5.014 20.142
Ingroup
Moral 0.357 1.400 224.467 1.000 ‐3.370 4.084
Immoral ‐7.674 4.634 68.080 0.614 ‐20.268 4.920
Outgroup ‐0.110 1.194 403.698 1.000 ‐3.277 3.057
Outgroup
Moral 0.467 1.319 245.921 1.000 ‐3.041 3.975
Immoral ‐7.564 4.628 67.760 0.641 ‐20.142 5.014
190
Ingroup 0.110 1.194 403.698 1.000 ‐3.057 3.277
Based on estimated marginal means
a. Dependent Variable: Corrugator Response Character Manipulation
b. Adjustment for multiple comparisons: Bonferroni.
Custom Hypothesis Test 1 (immoral linear vs ingroup linear)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval Lower Bound Upper Bound
L1 0.999 1.161 62.057 0.000 0.861 0.393 ‐1.322 3.321
a. immoral linear vs ingroup linear
b. Dependent Variable: corrugator response Character Manipulation
Custom Hypothesis Test 2 (ingroup linear vs outgroup linear)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval Lower Bound Upper Bound
L1 ‐1.068 0.664 225.174 0.000 ‐1.608 0.109 ‐2.377 0.241
a. ingroup linear vs outgroup linear
b. Dependent Variable: corrugator response Character Manipulation
Custom Hypothesis Test 3 (immoral linear vs outgrup linear)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval Lower Bound Upper Bound
L1 ‐0.069 1.208 84.431 0.000 ‐0.057 0.955 ‐2.471 2.333
a. immoral linear vs outgrup linear
b. Dependent Variable: corrugator response Character Manipulation
Custom Hypothesis Test 4 (outgroup quadratic vs immoral quadratic)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval Lower Bound Upper Bound
L1 ‐3.392 1.555 64.524 0.000 ‐2.181 0.033 ‐6.499 ‐0.286
a. outgroup quadratic vs immoral quadratic
b. Dependent Variable: corrugator response Character Manipulation
191
C3.3
Model Summary Affective State Adjective: Study 3
total cases
cases after baseline
data loss
Affective State Adjective Corrugator 79360 76620 3.45%
‐2 LL nr of parameters
p model fit (chi‐square distribution)
model comparision
added predictor action
Model 0 901497.227 2.0 Intercept Only
Model 1 897288.610 3.0 0.000 better Subject Random keep
Model 2 896040.156 4.0 0.000 better Item Random keep
Model 3 896039.477 5.0 0.410 not better Manipulation type Fixed keep for interaction
Model 4 896039.476 6.0 0.975 not better Likeability Fixed keep for interaction
Model 5 895996.454 7.0 0.000 better Event Valence Fixed keep
Model 6 895995.958 8.0 0.481 not better Manipulation Type*Likeability Fixed keep for interaction
Model 7 895993.409 9.0 0.110 not better Manipulation Type*Event Valence Fixed
keep for interaction
Model 8 895989.189 10.0 0.040 better Likeability*EventValence Fixed keep
Model 9 895982.633 11.0 0.010 better Manipulation Type*Likeability*Event Valence Fixed keep
Model 10 0.000 0.0 nb
no convergence
Manipulation Type*Likeability*Event Valence Random remove
Model 11 895982.629 12.0 0.950 not better Linear Moral Pos Fixed remove
Model 12 895826.012 12.0 0.000 better Linear Moral Neg Fixed keep
Model 13 895821.553 13.0 0.035 better Linear Immoral Pos Fixed keep
Model 14 895812.099 14.0 0.002 better Linear Immoral Neg Fixed keep
Model 15 895812.032 15.0 0.796 not better Linear Ingroup Pos Fixed remove
Model 16 895751.948 15.0 0.000 better Linear Ingroup Neg Fixed keep
Model 17 895751.244 16.0 0.401 not better Linear Outgroup Pos Fixed remove
Model 18 895691.052 16.0 0.000 better Linear Outgroup Neg Fixed keep
Model 19 893205.663 18.0 0.000 better Linear Moral Neg Random keep
Model 20 893049.734 21.0 0.000 better Linear Immoral Pos Random keep
Model 21 0.000 0.0 nb
no convergence Linear Immoral Neg Random remove
Model 22 892285.545 25.0 0.000 better Linear Ingroup Neg Random keep
Model 23 891728.969 30.0 0.000 better Linear Outgroup Neg Random keep
Model 24 891728.391 31.0 0.447 not better Quadratic Moral Pos Fixed remove
Model 25 891727.445 31.0 0.217 not better Quadratic Moral Neg Fixed remove
Model 26 891725.609 31.0 0.067 not better Quadratic Immoral Pos Fixed remove
Model 27 891727.977 31.0 0.319 not better Quadratic Immoral Neg Fixed remove
Model 28 891728.881 31.0 0.767 not better Quadratic Ingroup Pos Fixed remove
Model 29 891728.108 31.0 0.353 not better Quadratic Ingroup Neg Fixed remove
Model 30 891728.965 31.0 0.950 not better Quadratic Outgroup Pos Fixed remove
Model 31 891728.727 31.0 0.623 not better Quadratic Outgroup Neg Fixed remove
Model 32 891728.833 31.0 0.712 not better Cubic Moral Pos Fixed
Model 33 891722.749 31.0 0.013 better Cubic Moral Neg Fixed keep
Model 34 891721.970 32.0 0.377 not better Cubic Immoral Pos Fixed remove
Model 35 891722.532 32.0 0.641 not better Cubic Immoral Neg Fixed remove
Model 36 891722.494 32.0 0.614 not better Cubic Ingroup Pos Fixed
Model 37 891713.807 32.0 0.003 better Cubic Ingroup Neg Fixed keep
Model 38 891713.233 33.0 0.449 better Cubic Outgroup Pos Fixed remove
Model 39 891708.201 33.0 0.018 better Cubic Outgroup Neg Fixed keep
Model 40 0.000 34.0 nb
no convergence Cubic Moral Neg Random remove
192
Model 41 0.000 34.0 nb
no convergence Cubic Ingroup Neg Random remove
Model 42 0.000 34.0 nb
no convergence Cubic Outgroup Neg Random remove
Model 43 891706.713 34.0 0.223 not better Block Order remove
C3.4
Parameter Estimates Affective State Adjective: Study 3
Type III Tests of Fixed Effectsa Source Numerator df Denominator df F Sig. Manipulation Type*Likeability*Valence* 8.000 276.701 232.233 0.000 Linear Moral Neg 1.000 77.036 7.001 0.010 Linear Immoral Pos 1.000 62.226 1.118 0.294 Linear Immoral Neg 1.000 75797.107 10.031 0.002 Linear Ingroup Neg 1.000 118.517 11.055 0.001 Linear Outgroup Neg 1.000 150.251 11.745 0.001 Cubic Moral Neg 1.000 75797.107 6.221 0.013 Cubic Ingroup Neg 1.000 75797.107 8.944 0.003 Cubic Outgroup Neg 1.000 75797.107 5.607 0.018 a. Dependent Variable: Corrugator Response Affective State Adjective.
Estimates of Fixed Effectsa
Parameter Estimate Std. Error df t Sig.
95% Confidence Interval
Lower Bound Upper Bound
Moral Pos 109.164 3.215 128.000 33.953 0.000 102.803 115.526
Moral Neg 121.855 3.216 128.178 37.887 0.000 115.491 128.219
Immoral Pos 116.140 3.216 128.182 36.110 0.000 109.776 122.504
Immoral Neg 116.827 3.217 128.231 36.320 0.000 110.463 123.192
Ingroup Pos 110.187 3.215 127.935 34.275 0.000 103.825 116.548
Ingroup Neg 120.420 3.216 128.081 37.448 0.000 114.057 126.783
Ingroup Pos 108.677 3.216 128.089 33.795 0.000 102.314 115.040
Ingroup Neg 120.196 3.218 128.391 37.355 0.000 113.830 126.563
Linear Moral Neg 56.394 21.313 77.036 2.646 0.010 13.954 98.833
Linear Immoral Pos 6.593 6.235 62.226 1.057 0.294 ‐5.870 19.056
Linear Immoral Neg 9.113 2.877 75797.107 3.167 0.002 3.473 14.752
Linear Ingroup Neg 43.392 13.051 118.517 3.325 0.001 17.550 69.235
Linear Outgroup Neg 39.676 11.577 150.251 3.427 0.001 16.801 62.551
Cubic Moral Neg ‐117.162 46.973 75797.107 ‐2.494 0.013 ‐209.229 ‐25.096
Cubic Ingroup Neg ‐140.036 46.826 75797.107 ‐2.991 0.003 ‐231.815 ‐48.258
Cubic Outgroup Neg ‐111.869 47.246 75797.107 ‐2.368 0.018 ‐204.471 ‐19.268
a. Dependent Variable: Corrugator Response Affective State Adjective.
193
Custom Hypothesis Test (linear ingroupneg vs linear outgroupneg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 3.717 11.212 998.582 0.000 0.332 0.740 ‐18.285 25.718
a. linear ingroupneg vs linear outgroupneg
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (cubic ingroupneg vs cubic outgroupneg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐28.167 66.519 75797.107 0.000 ‐0.423 0.672 ‐158.544 102.210
a. cubic ingroupneg vs cubic outgroupneg
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (moral‐pos vs moral‐neg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐12.691 2.586 505.899 0.000 ‐4.907 0.000 ‐17.772 ‐7.609
a. moral‐pos vs moral‐neg
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (immoral‐pos vs immoral‐neg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐0.687 2.588 507.310 0.000 ‐0.265 0.791 ‐5.772 4.398
a. immoral‐pos vs immoral‐neg
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (ingroup‐pos vs ingroup‐neg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐10.234 2.585 504.923 0.000 ‐3.958 0.000 ‐15.313 ‐5.154
a. ingroup‐pos vs ingroup‐neg
194
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (ingroup‐pos vs outgroup‐pos)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 1.509 2.585 504.988 0.000 0.584 0.560 ‐3.570 6.588
a. ingroup‐pos vs outgroup‐pos
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (ingroup‐neg vs outgroup‐neg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 0.224 2.589 507.673 0.000 0.087 0.931 ‐4.862 5.310
a. ingroup‐neg vs outgroup‐neg
b. Dependent Variable: Corrugator Response Affective State Adjective.
Custom Hypothesis Test (outgroup‐pos vs outgroup‐neg)
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐11.519 2.589 507.743 0.000 ‐4.449 0.000 ‐16.605 ‐6.433
a. outgroup‐pos vs outgroup‐neg
b. Dependent Variable: Corrugator Response Affective State Adjective.
C3.5
Model Summary Affect Reason: Study 3
total cases
cases after baseline data loss
Affect Reason 198400 191550 3.45%
‐2 LL nr of parameters
p model fit (chi‐square distribution)
model comparision
Added predictor action
Model 0 2548720.526 2 Intercept Only
Model 1 2532885.727 3 0.000 better Subject Random keep
Model 2 2526668.46 4 0.000 better Item Random keep
195
Model 3 2526664.357 5 0.043 not better Manipulation type Fixed
keep for interaction
Model 4 2526663.69 6 0.414 not better Likeability Fixed keep for interaction
Model 5 2526523.61 7 0.000 better Event Valence Fixed keep
Model 6 2526523.154 8 0.499 not better Manipulation Type*Likeability Fixed
keep for interaction
Model 7 2526522.866 9 0.592 not better
Manipulation Type*Event Valence Fixed
keep for interaction
Model 8 2526510.725 10 0.000 better Likeability*EventValence Fixed keep
Model 9 2526505.938 11 0.029 better
Manipulation Type*Likeability*Event Valence Fixed keep
Model 10 0 0 n.b. no convergence
Manipulation Type*Likeability*Event Valence Random remove
Model 11 2526505.866 12 0.788 not better Linear Moral Pos Fixed remove
Model 12 2526328.258 12 0.000 better Linear Moral Neg Fixed keep
Model 13 2526321.29 13 0.008 better Linear Immoral Pos Fixed keep
Model 14 2526169.983 14 0.000 better Linear Immoral Neg Fixed keep
Model 15 2526169.979 15 0.950 not better Linear Ingroup Pos Fixed remove
Model 16 2526009.561 15 0.000 better Linear Ingroup Neg Fixed keep
Model 17 2526009.561 16 1.000 not better Linear Outgroup Pos Fixed remove
Model 18 2525950.654 16 0.000 better Linear Outgroup Neg Fixed keep
Model 19 2524008.527 18 0.000 better Linear Moral Neg Random keep
Model 20 2523705.108 21 0.000 better Linear Immoral Pos Random keep
Model 21 2521529.994 25 0.000 better Linear Immoral Neg Random keep
Model 22 2520054.253 30 0.000 better Linear Ingroup Neg Random keep
Model 23 2519616.072 36 0.000 better Linear Outgroup Neg Random keep
Model 24 2519616.036 37 0.850 not better Quadratic Moral Pos Fixed remove
Model 25 2519541.076 37 0.000 better Quadratic Moral Neg Fixed keep
Model 26 2519534.389 38 0.010 better Quadratic Immoral Pos Fixed keep
Model 27 2519506.896 39 0.000 better Quadratic Immoral Neg Fixed keep
Model 28 2519506.499 40 0.529 not better Quadratic Ingroup Pos Fixed remove
Model 29 2519451.237 40 0.000 better Quadratic Ingroup Neg Fixed keep
Model 30 2519451.047 41 0.663 not better Quadratic Outgroup Pos Fixed remove
Model 31 2519388.856 41 0.000 better Quadratic Outgroup Neg Fixed remove
Model 32 2518159.402 48 0.000 better Quadratic Moral Neg Random keep
Model 33 0 0 n.b. no convergence
Quadratic Immoral Pos Random remove
Model 34 0 0 n.b. no convergence
Quadratic Immoral Neg Random remove
Model 35 0 0 n.b. no convergence
Quadratic Ingroup Neg Random remove
Model 36 0 0 n.b. no convergence
Quadratic Outgroup Neg Random remove
Model 37 2518124.197 49 0.000 better Cubic Moral Neg Fixed keep
196
Model 38 2518120.8 50 0.065 not better Cubic Immoral Pos Fixed remove
Model 39 2518096.146 50 0.000 better Cubic Immoral Neg Fixed keep
Model 40 2518096.012 51 0.714 not better Cubic Ingroup Pos Fixed remove
Model 41 2518074.252 51 0.000 better Cubic Ingroup Neg Fixed keep
Model 42 2518074.233 52 0.890 no convergence
Cubic Outgroup Pos Fixed remove
Model 43 2518071.724 52 0.112 not better Cubic Outgroup Neg Fixed remove
Model 44 0 52 n.b. no convergence
Cubic Moral Neg Random remove
Model 45 0 52 n.b. no comvergence
Cubic Immoral Neg Random remove
Model 46 0 52 n.b. no convergence
Cubic Ingroup Neg Random remove
Model 47 2518072.146 52 0.147 not better Block Order remove
C3.5
Parameter Estimates Affective Reason: Study 3
Type III Tests of Fixed Effectsa Source Numerator df Denominator df F Sig. Manipulation Type*Likeability*Valence* 8 291.262 87.888 0.000 Linear Moral Neg 1 82.812 18.598 0.000 Linear Immoral Pos 1 61.735 1.012 0.318 Linar Immoral Neg 1 80.335 14.327 0.000 Linear Ingroup Neg 1 83.231 16.666 0.000 Linear Outgroup Neg 1 61.885 7.329 0.009 Quadratic Moral Neg 1 67.376 6.016 0.017 Quadratic Immoral Pos 1 190585.271 6.744 0.009 Quadratic Immoral Neg 1 190585.271 27.722 0.000 Quadratic Ingroup Neg 1 190585.271 56.109 0.000 Quadratic Outgroup Neg 1 190585.271 62.867 0.000 Cubic Moral Neg 1 190585.271 35.218 0.000 Cubic Immoral Neg 1 190585.271 28.057 0.000 Cubic Ingroup Neg 1 190585.271 21.895 0.000 a. Dependent Variable: Corrugator Response Affect Reason
Estimates of Fixed Effectsa
Parameter Estimate Std. Error df t Sig.
95% Confidence Interval
Lower Bound Upper Bound
Moral Pos 111.003 7.535 115.477 14.731 0.000 96.078 125.929
Moral Neg 172.344 7.639 121.958 22.561 0.000 157.222 187.466
Immoral Pos 129.447 7.639 121.958 16.946 0.000 114.325 144.569
Immoral Neg 151.351 7.639 121.988 19.812 0.000 136.227 166.474
Ingroup Pos 110.935 7.535 115.458 14.722 0.000 96.010 125.860
Ingroup Neg 161.521 7.638 121.887 21.147 0.000 146.401 176.641
Ingroup Pos 110.079 7.536 115.504 14.607 0.000 95.153 125.006
Ingroup Neg 153.927 7.641 122.099 20.144 0.000 138.800 169.053
Linear Moral Neg 41.989 9.736 82.812 4.313 0.000 22.623 61.355
Linear Immoral Pos 4.270 4.244 61.735 1.006 0.318 ‐4.214 12.755
197
Linar Immoral Neg 38.772 10.243 80.335 3.785 0.000 18.389 59.155
Linear Ingroup Neg 38.438 9.416 83.231 4.082 0.000 19.712 57.165
Linear Outgroup Neg 12.368 4.568 61.885 2.707 0.009 3.235 21.500
Quadratic Moral Neg ‐20.315 8.282 67.376 ‐2.453 0.017 ‐36.845 ‐3.785
Quadratic Immoral Pos ‐6.234 2.401 190585.271 ‐2.597 0.009 ‐10.939 ‐1.529 Quadratic Immoral Neg ‐12.653 2.403 190585.271 ‐5.265 0.000 ‐17.363 ‐7.943
Quadratic Ingroup Neg ‐17.926 2.393 190585.271 ‐7.491 0.000 ‐22.616 ‐13.235 Quadratic Outgroup Neg ‐19.145 2.415 190585.271 ‐7.929 0.000 ‐23.877 ‐14.412
Cubic Moral Neg ‐22.639 3.815 190585.271 ‐5.934 0.000 ‐30.116 ‐15.162
Cubic Immoral Neg ‐20.228 3.819 190585.271 ‐5.297 0.000 ‐27.712 ‐12.743
Cubic Ingroup Neg ‐17.794 3.803 190585.271 ‐4.679 0.000 ‐25.248 ‐10.341
a. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 34.501 8.085 95.646 0.000 4.267 0.000 18.451 50.552
a. linear immoral‐pos vs. immoral‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐26.071 6.412 124.783 0.000 ‐4.066 0.000 ‐38.760 ‐13.381
a. linear ingroup‐neg‐neg vs. outgroup‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐1.219 3.400 190585.271 0.000 ‐0.359 0.720 ‐7.882 5.444
a. quadratic ingroup‐neg‐neg vs. outgroup‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐61.340 5.755 554.539 0.000 ‐10.658 0.000 ‐72.645 ‐50.036
a. moral‐pos vs moral‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐18.444 5.755 554.546 0.000 ‐3.205 0.001 ‐29.748 ‐7.139
198
a. moral‐pos vs immoral‐pos
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 20.993 5.891 608.660 0.000 3.564 0.000 9.424 32.562
a. moral‐neg vs immoral‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐21.904 5.891 608.663 0.000 ‐3.718 0.000 ‐33.472 ‐10.335
a. immoral‐pos vs immoral‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐50.586 5.753 553.830 0.000 ‐8.792 0.000 ‐61.887 ‐39.285
a. ingroup‐pos vs ingroup‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 0.856 5.617 503.241 0.000 0.152 0.879 ‐10.180 11.892
a. ingroup‐pos vs outgroup‐pos
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 7.594 5.892 608.997 0.000 1.289 0.198 ‐3.976 19.164
a. ingroup‐neg vs outgroup‐neg
b. Dependent Variable: Corrugator Response Affect Reason
Contrast Estimatesa,b
Contrast Estimate Std. Error df Test Value t Sig.
95% Confidence Interval
Lower Bound Upper Bound
L1 ‐43.847 5.759 555.900 0.000 ‐7.614 0.000 ‐55.159 ‐32.536
a. outgroup‐pos vs outgroup‐neg
b. Dependent Variable: Corrugator Response Affect Reason
199
Supplementary Information D
D1 – Study 1
D1 – Study 1
200
D2 – Study 2
D2 – Study 2
201
D2 – Study 2
202
D3 – Study 3
203
D3 – Morality: Study 3
D3 – Minimal Groups: Study 3
204
D3 – Morality: Study 3
D3 – Minimal Groups: Study 3
205
Supplementary Information E
E1 ‐ Study 2
In our previous study ('t Hart B. , Struiksma, van Boxtel, & Van Berkum, Emotion in
Stories: A facial EMG study on simulation vs. moral evaluation, under review) we
found subtle differences in the way the corrugator responded to the manipulation
of character morality and affective event valence based on scores from the
Adolescent Measure of Empathy and Sympathy (Vossen, Piotrowski, & Valkenburg,
Development of the Adolescent Measure of Empathy and Sympathy (AMES), 2015).
This questionnaire measures three components of pro‐social emotion: cognitive
empathy (understanding what the other feels), affective empathy (feeling what the
other feels), and sympathy (feeling for the other). In our previous experiment we
found that the higher participants scored on Affective Empathy the more negative
affect the corrugator reflected at immoral actions and at unfair affective events
(moral‐negative and immoral‐positive).
We have also included the Moral Foundations Questionnaire (Graham, et
al., 2011) to investigate whether participants’ attachment to the morality
dimension of fairness influenced the corrugator response. We suspected fairness to
be an especially salient moral dimension because the evaluation of the affective
event hinges on whether it is considered fair or unfair in light of the protagonist’s
moral status.
We calculated an average score for each subscale of the AMES for each
subject, ranging from 1 to 5. To investigate the effect of each of these individual
measures on the fEMG response, we used the mixed models procedure. We
analysed individual differences in average corrugator activation to each of the
three segments by including each personality trait as a continuous covariate in the
fixed part of a simplified model without time components. The personality traits
included affective empathy (M = 3.04, range = 1.25‐4.25), cognitive empathy (M =
206
3.56, range = 2.25‐5.00), sympathy (M = 3.98, range = 2.33‐5.00) and fairness (M =
3.66, range = 2.50‐4.80).
At the character morality segment we found an effect for affective
empathy only. Our analysis revealed that if participants scored higher on the
affective empathy scale, they tended to frown less in response to immoral actions
(b = −26.98, t (60.22) = −2.52, p = .014, 95% CI [−48.39, −5.59]). This result not only
did not replicate our finding from the previous experiment, it actually displayed the
exact reverse effect; higher affective empathy scores led to more rather than less
frowning during immoral actions.
At the affective state adjective segment corrugator activity only interacted
with cognitive empathy, and there only for the moral‐negative. The effect was such
that the higher participants scored on cognitive empathy, the less they frowned at
bad things happening to good people (B = −11.73, t (65.17) = −2.81, p < .01, 95% CI
[−20.07, −3.40]). Higher cogni ve empathy scores indicate a be er developed skill
at understanding how people feel. Of course, a better understanding of how
someone feels does not necessarily mean that you will feel bad for them
(sympathetic response) and feel what they feel (affective empathetic response).
This effect of cognitive empathy was new, we found no such effects in our previous
study, although there we were unable to test the effect of these traits on the
affective state adjective separately.
At the affect reason segment we found an effect of three traits for moral‐
negative, i.e., bad things happening to good people. First, we once again found an
effect of cognitive empathy for moral‐negative. This effect was in the same
direction as for the affective state adjective segment: higher cognitive empathy
scores meant less corrugator activity (B = −26.03, t (63.38) = −4.36, p < .001, 95% CI
[−37.98, −14.09]).
Secondly, increasing affective empathy scores corresponded to a decrease
of negative affect on the corrugator. This result is similar to that during the
character morality segment (B = −10.96, t (63.92) = −2.05, p = .045, 95% CI [−21.65,
207
−0.27]). Affective empathy measures the tendency of people to feel what another
feels, we would therefore expect more negative affect for good characters meeting
misfortune for participants who scored higher on this trait. In fact, we found
precisely that previously ('t Hart B. , Struiksma, van Boxtel, & Van Berkum, Emotion
in Stories: A facial EMG study on simulation vs. moral evaluation, under review).
The result here represents is the opposite.
The third individual trait that revealed an effect on corrugator activity in
response to moral‐negative conditions was fairness. This measure reflected how
strongly a person adhered to the moral dimension of fairness. Here the effect
seemed more expected; the more people adhered to fairness, the more they
frowned at bad things happening to good people, which is unfair (B = 28.42, t
(65.04) = 4.65, p < .001, 95% CI [16.22, 40.62]). However, higher scores for fairness
also led to more frowning in response to immoral‐negative conditions, which might
be considered fair as it concerns bad characters experiencing something bad (B =
13.92, t (65.11) = 2.28, p = .026, 95% CI [1.71, 26.14]).
All in all, the results of the individual differences present a mixed bag. We
find some effects that go against not only our expectations, but also the results
obtained in a previous experiment with near enough the same materials. For this
reason, as well as the correlational nature of these results, we are wary of drawing
strong conclusions based on these analyses. While we maintain that these
processes of simulation and evaluation could be subject to some individual
variation, our conflicting results as well as the correlational nature of the evidence
do not give us a solid foundation to make any claims regarding individual
differences at this time.
208
E2 – Study 3
Participants completed two questionnaires after completing the facial EMG part of
the experiment: the Adolescent Measure of Empathy and Sympathy (AMES)
(Vossen, Piotrowski, & Valkenburg, 2015) and the Moral Foundations
Questionnaire (MFQ) (Graham, et al., 2011).The AMES measures traits on a 5‐point
scale related to pro‐social emotions and makes a useful distinction between
affective empathy, i.e., feeling what another feels, (M = 3.13, range = 1.5‐4.75) and
cognitive empathy, i.e., understanding what another feels, (M = 3.64, range = 2.5‐
4.75) as well as measuring sympathy, i.e., feeling for another (M = 4.02, range =
3.00‐5.00). The MFQ measures the importance an individual places on several
dimensions of morality. We focused our analysis on the moral dimension of
fairness as we consider this to be the most relevant dimension in relation to the
evaluation of the affective states and events described for the different characters.
For instance, it may be unfair for a good/ingroup character to experiencing
something bad, but potentially less unfair if the same is experienced by a
bad/outgroup character. The fairness subscale is measured using 6 questions on a
5‐point scale and adding the answers to produce a total score (M = 21.73, range =
13.00‐29.00)
To investigate the effects of these individual difference variables we once
again employed the mixed models procedure, but without the growth‐curve model
to investigate the temporal development of the corrugator signal. Instead, we
tested the potential effect of each individual difference variable on the average
activation for each of our experimental conditions We tested this separately for the
three critical segments: character manipulation, affective state adjective, and affect
reason: e.g., does a higher affective empathy score lead to higher average
corrugator activity while processing negative adjectives describing the emotional
state of a moral ingroup character?
209
Character Manipulation.
For the segment containing the description of the character as either moral
in/outgroup or minimal in/outgroup, scores for affective empathy, sympathy, and
fairness revealed a significant interaction with the overall average corrugator
activity for immoral characters only. In the case of affective empathy we found that
the higher the score for affective empathy was, the more frowning we measured
when presented with a moral outgroup character (b = 6.80, t (65.91) = 3.60, p <
.001, 95% CI [3.03, 10.58]). This indicated that those who are more inclined to feel
what another feels, experienced more negative affect when confronted, however
abstractly, with a ‘bad’ person. The interaction with sympathy went in the same
direction, the higher the sympathy score, the more people frowned in response to
being introduced to a moral outgroup character (b = 12.28, t (66.00) = 5.97, p <
.001, 95% CI [8.18, 16.39]). This pointed to a tendency to experience more negative
affect by those participants that were more inclined to feel for someone. The final
interaction for this segment described the effect that those who attached a greater
importance to fairness tended to frown less in response to being presented with a
character described as moral outgroup b = −1.81, t (65.54) = −5.74, p < .001, 95% CI
[−2.44, −1.18]).
Affective State Adjective.
During this segment the characters were described to experience a positive or
negative emotion. We once again found interactions with several conditions for
affective empathy and sympathy, but not for fairness. We found that affective
empathy scores led to higher average activation of the corrugator for both moral
ingroup‐negative (b = 18.88, t (75.04) = 4.00, p < .001, 95% CI [9.47, 28.29]) and
moral outgroup‐positive (b = 15.05, t (75.16) = 3.19, p < .01, 95% CI [5.64, 24.46]).
We found that higher affective empathy scores also led to higher average
activation for minimal ingroup‐negative (b = 18.79, t (74.95) = 3.98, p < .001, 95% CI
[9.39, 28.20]) and minimal outgroup‐negative (b = 14.53, t (75.92) = 3.07, p < .01,
210
95% CI [5.10, 23.96]). These results are intriguing because while these conditions
contain different valence affective state adjectives, they were all evaluated as
unfair when you view the valence of the affective state in light of the moral status
of the character. In fact, the main results revealed that the corrugator response
reflected a negative evaluation for all these conditions.
We also found effects of sympathy during this segment. The conditions
that revealed interactions with sympathy were all those that contained a negative
affective state adjective. For the morality‐based conditions we found that higher
sympathy scores led to significantly higher average corrugator activity for moral
ingroup‐negative (b = 32.20, t (74.92) = 6.21, p < .001, 95% CI [21.86, 42.53]) and
moral outgroup‐negative (b = 14.99, t (75.20) = 2.89, p < .01, 95% CI [4.65, 25.33]).
Higher sympathy scores also led to more corrugator activity for minimal ingroup‐
negative (b = 12.68, t (74.96) = 2.45, p = .017, 95% CI [2.35, 23.00]) and minimal
outgroup‐negative (b = 15.63, t (75.45) = 3.01, p < .01, 95% CI [5.29, 25.97]). Taken
together, those participants with a greater tendency to ‘feel what others feel’
displayed more negative affect in response to negative emotions ascribed to any
character, regardless of in/outgroup status. This provides support for the general
conclusion that contextual evaluation plays an important role in how affectively
salient language is processed online.
Affect Reason
In this segment participants read the final part of the sentence that supplied the
last bit of context for the affective state the characters were in: the reason or the
events that elicited the affective state. We once again found interactions for both
affective empathy and sympathy scores, but also, for one condition, for cognitive
empathy. Higher affective empathy scores led to increased average corrugator
activity for moral ingroup‐negative (b = 27.99, t (65.32) = 2.33, p = .023, 95% CI
[4.05, 51.93]), minimal ingroup‐negative (b = 28.99, t (65.29) = 2.42, p = .018, 95%
211
CI [5.05, 52.93]), and minimal outgroup‐negative (b = 29.61, t (65.53) = 2.47, p =
.016, 95% CI [5.65, 53.65]).
Higher sympathy scores again led to increased frowning activity for all four
negative conditions: moral ingroup‐negative (b = 63.13, t (65.43) = 4.88, p < .001,
95% CI [37.31, 88.94]), moral outgroup‐negative (b = 26.00, t (65.50) = 2.01, p =
.049, 95% CI [0.17, 51.83]) minimal ingroup‐negative (b = 36.02, t (65.43) = 2.79, p <
.01, 95% CI [10.20, 61.84]), and minimal outgroup‐negative (b = 40.47, t (65.57) =
3.13, p < .01, 95% CI [14.65, 66.29]). We also found an effect of cognitive empathy
here for the first time. Higher scores for cognitive empathy led to more frowning
for moral ingroup‐negative conditions (b = 39.02, t (65.65) = 2.43, p = .018, 95% CI
[7.00, 71.04]).
These results are highly suggestive, in particular the blanketing result that
higher sympathy corresponded to more negative affect being reflected in overall
corrugator activity in response to both negative state adjectives and negative affect
reasons, regardless of grouping dimension or in/outgroup status. These findings
run contrary to previous studies, but this could be due to the reduced impact of the
propositional character manipulation here, compared to the narrative
manipulation in previous studies. Higher affective empathy scores also reliably lead
to increased corrugator activity during all negative state adjectives and reasons.
These results varied only for those conditions where the main corrugator results
revealed the clearest conflict of language‐driven and simulation: moral outgroup‐
positive and moral outgroup‐negative.
Having said that, these results are correlational and also vary considerably
from the effects we observed in previous studies using the same or similar
materials. We therefore report these data as interesting, but refrain from
formulating concrete conclusions based on them.
212
Samenvatting in het Nederlands
Dit proefschrift onderzoekt de invloed van twee processen tijdens het verwerken
van affectieve talige stimuli; evaluatie en taal‐gestuurde simulatie. Met simulatie
wordt hier bedoeld het opnieuw activeren van perceptuele, motorische, en
introspectieve (inclusief affectieve) sporen van eerdere ervaringen met de referent
van de talige stimuli die verwerk worden (Barsalou, 2008). Dit proces van taal‐
gestuurde simulatie wordt binnen het kader van gegronde cognitie als (al dan niet
noodzakelijk) onderdeel van taalbegrip gezien. Met evaluatie wordt hier bedoeld
de affectieve respons gebaseerd op de evaluatie van zowel individuele talige
stimuli als de situatie waaraan de talige stimulus als geheel refereert (zie Van
Berkum, 2017).
Een aanzienlijk aantal studies heeft met behulp van elektromyografie
(EMG) aangetoond dat het verwerken van affectieve, talige stimuli (vaak losse
zelfstandige‐ of bijvoeglijke naamwoorden of korte zinnen) inderdaad gepaard gaat
met congruente activatie van gezichtsspieren waarvan het bekend is dat deze
gelinkt zijn aan emotie. De corrugator superciliii, of de ‘fronsspier’, wordt vaak
gebruikt in deze studies vanwege de zeer betrouwbare correspondentie tussen de
activatiepatronen van deze spier en affectieve valentie, d.w.z. positief vs. negatief.
De corrugator neemt toe in activiteit in respons op negatieve stimuli en neemt af in
respons op positieve stimuli. Dit soort congruente gezichtsspier activiteit wordt
vaak geïnterpreteerd als taal‐gestuurde simulatie als onderdeel van taalverwerking.
Echter, dit type experiment maakt vaak gebruik van ondubbelzinnig
positieve en negatieve talige stimuli. In het dagelijks leven komen we veel taal
tegen waar de affectieve valentie niet zo ondubbelzinnig is. We evalueren, als
taalverwerker, vaak ook nog iets van wat we lezen of horen. Een interessant
voorbeeld waar simulatie en evaluatie van één en dezelfde stimulus niet dezelfde
valentie hebben is te vinden in narratieven. Narratieven gaan vaak over karakters
waar we iets van vinden. De evaluatie van het karakter kan invloed hebben op hoe
213
affectieve talige stimuli met betrekking tot datzelfde karakter worden verwerkt.
Neem de volgende zin:
1. Mark is verdrietig als hij ziet dat zijn gloednieuwe laptop kapot is gevallen
Afhankelijk van wat de lezer van een dergelijke zin van Mark vindt zou je kunnen
voorspellen dat simulatie en evaluatie wel of niet met elkaar overeen zouden
komen. In het geval dat de lezer Mark een slecht mens vindt zou de evaluatie van
zijn verdriet over zijn laptop eerder positief uit kunnen pakken, terwijl taal‐
gestuurde simulatie alsnog een negatieve corrugator respons voorspelt. Dit
proefschrift stelt drie verschillende modellen op voor hoe deze twee processen uit
zouden pakken tijdens online taalverwerking.
Als eerste, enkel‐simulatie. Dit model voorspelt dat, ongeacht de evaluatie
van de affectieve betekenis van talige stimuli op basis van wat de lezer van het
karakter vindt, de corrugator altijd taal‐gestuurde simulatie zal reflecteren. Het
tweede model, evaluatie‐blokkeert‐simulatie, voorspelt dat de corrugator wellicht
beschikbaar is voor taal‐gestuurde simulatie in ondubbelzinnige en
ongecompliceerde situaties, maar dat zodra de lezer affectief betrokken is bij de
talige stimuli alleen evaluatie nog zichtbaar zal zijn in de corrugator activiteit. Het
derde model, gecombineerde‐aandrijving, voorspelt dat zowel evaluatie als
simulatie bijdragen aan het patroon van corrugator activiteit tijdens online
taalverwerking. Deze drie modellen worden in een serie van drie experimenten
getoetst.
Experiment 1
In experiment 1 lazen deelnemers korte verhaaltjes (zie tabel 2.1) die op twee
kritieke segmenten zijn gemanipuleerd om evaluatie en simulatie tegenover elkaar
te zetten. In het eerste kritieke segment was te lezen hoe de hoofdpersoon in het
214
verhaal moreel of immoreel handelde in dezelfde situatie. Deze manipulatie had
een duidelijk differentieel corrugator patroon tot gevolg, waarbij immorele
handelingen een aanzienlijke toename in corrugator activiteit tot gevolg had en
morele handelingen een meer bescheiden afname in activiteit. Dit resultaat toonde
aan dat de gevoeligheid van de corrugator voor affectieve valentie ook strekt tot
valentie op basis van moraliteit. Daarnaast bevestigde dit patroon dat onze
manipulatie succesvol was en dus de voorwaarden voor een contrasterend effect
van simulatie en evaluatie bij het volgende segment voldaan waren.
Het tweede kritieke segment beschreef een positieve of negatieve
affectieve episode voor de hoofdpersoon zoals in voorbeeld 1 hierboven. Hier
zagen we een duidelijk effect van de morele status van de hoofdpersoon. Voor
morele karakters zagen we een duidelijk differentieel patroon van corrugator
activiteit: toenemende activiteit voor negatieve affectieve episodes, en afnemende
activiteit voor positieve episodes. Dit patroon strookte met eerdere studies, maar
voor deze condities voorspelden zowel simulatie als evaluatie een dergelijke
patroon.
In het geval van immorele karakters voorspelden evaluatie en simulatie
tegenovergestelde patronen. Een slecht karakter dat iets goeds overkomt, zou als
negatief worden geëvalueerd, terwijl simulatie in een dergelijk geval nog steeds
een positieve respons zou voorspellen. De corrugator resultaten lieten zien dat in
het geval van immorele karakters, er in het geheel geen differentieel patroon
ontstond. Positieve als negatieve affectieve episodes van immorele karakters
ontlokten dezelfde corrugator respons zonder duidelijk toename of afname.
Deze resultaten strookten niet met het enkel‐simulatie model of het
evaluatie‐blokkeert‐simulatie model. De voorlopige conclusie was dan ook dat
gecombineerde‐aandrijving het juiste model was, waarbij de tegengestelde
krachten van simulatie en evaluatie in het geval van immorele karakters elkaar
ophieven.
215
Experiment 2
In experiment twee hebben we de stimuli verfijnd. Waar in experiment 1 het twee
kritieke segment, de affectieve episode, in één blok werden gepresenteerd, werden
hier de saillante affectieve elementen uit elkaar gehaald om de corrugator respons
beter te kunnen evalueren in relatie tot de affectieve informatie in het segment
(zie tabel 3.1).
2. Mark / is / verdrietig / als hij ziet dat / zijn gloednieuwe laptop kapot is
gevallen
Op deze manier hoopten we eventuele corrugator responses die door het gebrek
aan precisie in experiment 1 uitgesmeerd waren over de gehele tijdsperiode, uit
elkaar te kunnen trekken. Verder waren de stimuli identiek.
Wederom vonden we een duidelijk differentieel patroon van corurgator
activiteit tijdens het eerste kritieke segment; de karakter manipulatie. Wederom
ontlokten immorele acties een aanzienlijke toename in corrugator activiteit en
zagen we een numerieke afname als reactie op morele acties. Dit is niet alleen een
replicatie van de gevoeligheid van de corrugator voor morele valentie, maar gaf
ook wederom aan dat de manipulatie succesvol was.
Het eerste kritieke segment uit de affectieve episode was altijd een
bijvoeglijk naamwoord dat een affectieve staat beschreef van het karakter. Hier
zagen we wederom een duidelijk differentieel patroon ontstaan voor morele
karakters: toenemende activiteit voor negatieve emotionele toestanden en
afnemende activiteit voor positieve emotionele toestanden. Echter zagen we voor
immorele karakters, net als in het eerste experiment, geen differentieel patroon
voor positieve en negatieve emotionele toestanden. Wat dit segment betreft was
dit een replicatie van eerdere resultaten.
Het tweede kritieke segment uit de affectieve episode liet een iets ander
beeld zien. Hier zagen we voor zowel morele als immorele karakters een
216
differentieel patroon in de verwachte richting. Echter, het differentiële patroon
voor immorele karakters was verminderd in vergelijking met dat voor morele
karakters. Al met al leek dit experiment wederom aan te tonen dat het
gecombineerde‐aandrijving model het juiste was, met als verfijning dat de balans
van simulatie en evaluatie kan verschuiven afhankelijk van het type affectieve
informatie; d.w.z. een bijvoeglijk naamwoord dat een emotionele toestand
beschrijft vs. een frase die een affectief saillante gebeurtenis omschrijft.
Experiment 3
In het derde experiment probeerden we de juistheid van het gecombineerde‐
aandrijving model aan te tonen door te laten zien dat de mate van evaluatie te
manipuleren valt, en zo dus ook de uitkomst van de tegengestelde krachten van
evaluatie en simulatie. Dit deden we door een manipulatie van het karakter op
basis van moraliteit te contrasteren met een manipulatie op basis van een redelijk
nietszeggende, minimale groepen manipulatie. Minimale groepen zijn gebaseerd
op een betekenisloze classificatie van de proefpersoon. Vervolgens lazen de
proefpersonen dus over morele en immorele, maar ook over karakters die ofwel
tot dezelfde minimale groep behoorden (ingroep), of tot een andere minimale
groep (uitgroep).
Bij dit experiment werd de karaktermanipulatie niet langer bewerkstelligd
door een verhalende intro, maar met een simpele vaststelling:
3. Mark is een goed/slecht persoon
4. Mark is een type P/ type O persoon
De corrugator respons tijdens dit segment waren interessant omdat we voor de
moraliteitsmanipulatie slechts een trend‐niveau respons zagen en voor de
minimale groepen in het geheel geen reactie. Dit laat zien dat de evaluatieve
impact van moraliteit inderdaad groter is dan van minimale groepen. Daarnaast
217
laat het ook zien het beschrijven van een concrete situatie als manipulatie meer
impact heeft dan het presenteren van die manipulatie als propositie. Dit is de
kracht van narratief.
De kritieke segmenten van de affectieve episode waren hetzelfde als in
voorbeeld 2. Voor de moraliteitsmanipulatie vonden we, verrassend genoeg, bij het
bijvoeglijk naamwoord dat een emotionele toestand omschreef wederom een
differentieel patroon voor morele karakters (aanzienlijke toename voor negatieve
staten en lichte afname voor positieve staten) en geen verschil voor immorele
karakters (niet significante toename voor zowel positieve als negatieve staten). Dit
is verrassend omdat hieruit blijkt dat dit effect ook op kan treden zonder dat
daaraan een duidelijke affectieve respons op het karakter vooraf gaat. Voor de
minimale groepen vonden we geen enkel verschil voor in‐ en uitgroep condities;
beide lieten een duidelijk differentieel patroon zien voor positieve (lichte afname)
en negatieve (duidelijke toename) toestanden.
Het laatste kritieke segment, de frase die een affectief saillante
gebeurtenis beschrijft, liet voor de moraliteitscondities wederom een verminderd
differentieel effect zien voor immorele karakters ten opzichte van morele
karakters. Verrassend genoeg vonden we hier ook een licht verminderd
differentieel patroon voor uitgroep karakters ten opzichte van ingroep karakters.
We zijn voorzichtig met het interpreteren van dit laatste resultaat, omdat dit
slechts de eerste keer dat we dit aantonen. De resultaten voor de
moraliteitscondities, echter, repliceren resultaten van de eerdere experiment.
De drie studies tesamen bieden duidelijk bewijs tegen het intepreteren van
corrugator activiteit tijdens online taalverwerking in termen van simulatie alleen.
Simulatie alleen kan de keer op keer gerepliceerde patronen niet verklaren. Een
verklaring in termen evaluatie alleen lijkt eveneens niet afdoende gezien het
uitblijven van een evaluatieve respons op stimuli die doorgaans wel degelijk
evaluatie oproepen (denk aan het lezen van romans en roddelen over derden). Het
218
lijkt er dus op dat, van de drie a priori opgestelde modellen, het gecombineerde‐
aandrijving model het meest plausibel is. Echter, alternatieven als een model
waarin verschillende typen evaluatie tegen elkaar in werken (bijvoorbeeld
empathie en leedvermaak of gerechtigdheid) of een hergeformuleerde definitie
van taal‐gestuurde simulatie waarin voorgaande informatie een beperkend effect
heeft op simulatie van latere segmenten. In ieder geval zal toekomstig onderzoek
naar affectieve taalverwerking binnen een raamwerk van gegronde cognitie
rekening moeten houden met de complexiteit van affectieve talige stimuli in
termen van, in ieder geval, evaluatie en simulatie.
219
Curriculum Vitae
Björn ‘t Hart was born on October 9th, 1984 in Assen, the Netherlands. After
obtaining his VWO diploma in 2004, he studied English Language and Culture at
Groningen University. In 2011, he graduated with a Bachelor’s degree, specialising
in English literature. He went on to study English Linguistics at Leiden University,
obtaining a Master’s degree (cum laude) in 2012.
This dissertation is the result of work carried out between 2012 and 2017
as a PhD researcher at Utrecht University, within the NWO‐funded project ‘Moving
the language user – Affect and perspective in discourse processing’. Björn ‘t Hart
currently works as a lecturer with the department of Communication and
Information Sciences as Utrecht University.