The Emotional Embodiment of Words
Naveed A. Sheikh
Department of Psychology, Faculty of Science
McGill University
Montreal, Quebec, Canada
September 2015
A thesis submitted to McGill University in partial fulfillment of the requirements for the degree
of Doctor of Philosophy in Experimental Psychology
© Naveed A. Sheikh 2015
TABLE OF CONTENTS
TABLE OF CONTENTS .............................................................................................................. 2
ACKNOWLEDGEMENTS ......................................................................................................... 6
ABSTRACT ................................................................................................................................... 7
ABRÉGÉ ...................................................................................................................................... 10
FUNDING .................................................................................................................................... 14
STATEMENT OF ORIGINALITY ........................................................................................... 15
CONTRIBUTION OF AUTHORS ............................................................................................ 17
CHAPTER 1: ............................................................................................................................... 19
GENERAL INTRODUCTION .................................................................................................. 19
Embodied Word Representations .............................................................................................. 22
Direct Representation in Sensorimotor Vs. Emotional Experience .......................................... 27
L1 Emotional Word Processing ................................................................................................. 35
L2 Emotional Word Processing ................................................................................................. 38
Eye-Movement Measures of Reading as a Means of Investigating Embodiment ..................... 41
The Present Dissertation ........................................................................................................... 44
CHAPTER 2: ............................................................................................................................... 49
Sensorimotor and Linguistic Information Attenuate Emotional Word Processing Benefits: .... 49
An Eye-Movement Study............................................................................................................ 49
Abstract .................................................................................................................................... 50
Introduction ............................................................................................................................. 51
The Present Study .................................................................................................................. 55
Method ..................................................................................................................................... 57
Participants ........................................................................................................................... 57
Materials and Design ............................................................................................................ 57
2
Apparatus and Procedure ..................................................................................................... 61
Results ...................................................................................................................................... 61
Gaze Duration ....................................................................................................................... 63
Second Pass Time .................................................................................................................. 67
Fixation Probability and Regression Probability ................................................................. 67
Interactions with Alexithymia ............................................................................................... 68
Discussion................................................................................................................................. 72
PREFACE FOR CHAPTER 3 ................................................................................................... 81
CHAPTER 3: ............................................................................................................................... 83
The Embodiment of Emotional Words in a Second Language: An Eye-Movement Study. . 83
Abstract .................................................................................................................................... 84
Introduction ............................................................................................................................. 85
The Present Study .................................................................................................................. 89
Method ..................................................................................................................................... 90
Participants ........................................................................................................................... 90
Materials and Design ............................................................................................................ 90
Apparatus and Procedure ..................................................................................................... 92
Results ...................................................................................................................................... 92
L2 Readers vs. L1 Readers.................................................................................................... 98
Discussion................................................................................................................................. 99
PREFACE FOR CHAPTER 4 ................................................................................................. 103
CHAPTER 4: ............................................................................................................................. 105
The parafoveal and foveal effects of negative and positive emotional embodiment: ......... 105
Evidence from eye movements and the gaze-contingent boundary paradigm .................... 105
Abstract .................................................................................................................................. 106
3
Introduction ........................................................................................................................... 107
The Present Study .................................................................................................................112
Method ....................................................................................................................................114
Participants ..........................................................................................................................114
Materials ..............................................................................................................................114
Apparatus And Procedure ................................................................................................... 121
Results .................................................................................................................................... 121
Early Reading Measures on the Verb .................................................................................. 125
Early Reading Measures on the Noun ................................................................................ 128
Total Reading Time On The Noun ....................................................................................... 131
Fixated Definite Article ....................................................................................................... 132
Discussion............................................................................................................................... 133
Foveal Processing ............................................................................................................... 135
Parafoveal Processing ........................................................................................................ 140
CHAPTER 5: ............................................................................................................................. 146
GENERAL DISCUSSION........................................................................................................ 146
Summary of Studies ................................................................................................................. 148
Emotional Embodiment In Native Speakers ........................................................................... 157
Emotional Embodiment in L2 Speakers .................................................................................. 161
Embodiment Based on Negative vs. Positive Emotion............................................................ 165
Directions for Future Research ............................................................................................... 172
Conclusion .............................................................................................................................. 177
R E F E R E N C E S .................................................................................................................... 178
APPENDICES ........................................................................................................................... 199
Appendix A ............................................................................................................................. 200
4
Appendix B ............................................................................................................................. 206
5
ACKNOWLEDGEMENTS
I would like to extend my sincerest thanks to a number of people who have supported me
throughout graduate school. First and foremost, I would like to acknowledge my supervisor Dr.
Debra Titone. Debra's guidance over the past six years has enabled me to grow as a researcher.
Debra cultivated and nurtured the ideal environment for her students in her lab, one that
encourages both the free flow of ideas and academic rigor. She has gone above and beyond in
terms of the encouragement and support that she provided me throughout my tenure at McGill
University, and her dedication to her students is unparalled. I will be forever grateful for her
guidance and generous support.
I would like to thank my committee members Fred Genesee and Jelena Ristic.
Discussions with them in committee meetings have always been a joy, and I am thankful for their
many insightful comments on my research. I would like to thank the members of Debra's
laboratory, past and present, for their support over the years. In particular, I would like to thank
Irina Pivneva, Veronica Whitford, and Kyle Lovseth. They made the lab a wonderful space that
encouraged friendship and learning and paved the way for many of the lab's successes.
I would like to thank the many undergraduate honor's students, research assistants, and
work study students that I had the pleasure of working with over the years. This research would
not have been possible without their support and effort. I would also like to acknowledge
financial support from the Natural Research and Engineering and Research Council, the fonds
québécois de la recherche sur la société et la culture, and the fonds québécois de la recherche sur
la nature et les technologies.
I would also like to thank the many people who supported me personally. In particular, I
would like to thank my parents, Rahila and Nazir, who taught me the value of determination.
Finally, I would like to thank Sandra, the love of my life, who has shared in all of my
accomplishments and struggles.
6
ABSTRACT
The traditional view in cognitive science holds that word meanings are represented by
abstract symbols in encapsulated cognitive modules that cannot directly access modality-specific
information. This view continues to influence theories of word processing. For example, formal
models of eye movements during reading (e.g., Reichle, Rayner, & Pollatsek, 2003) generally
omit the potential role played by modality-specific experiences in word recognition (see also
Kuperman, Estes, Brysbaert, & Warriner, 2014). Over the last twenty years, this traditional view
has been increasingly challenged by an alternative theoretical approach which proposes that
word meanings are directly represented in sensory and motor systems. This alternative
theoretical approach is known as an embodied approach to cognition because it proposes that
word meanings are embodied or “grounded” in experiential knowledge (as opposed to abstract
symbols, as in the traditional view).
However, an often cited problem with embodied theories of cognition concerns how to
ground intangible abstract concepts that cannot be directly experienced. For example, the word
justice has an abstract referent that cannot be directly experienced. To address this question,
Vigliocco, Meteyard, Andrews, and Kousta (2009) proposed that words with concrete tangible
referents (e.g., table) are grounded in sensorimotor experiences, and words with abstract
intangible referents (e.g., justice) are grounded in emotional experiences, in addition to being
represented by linguistic information (e.g., word associations). In this dissertation, I test the idea
of emotional embodiment in word representation.
The first study (Chapter 2) investigates the idea that emotional facilitation of word
processing may not depend solely on words’ emotionality. Drawing on an embodied approach to
word representation, I examined interactions between emotional, sensorimotor, and linguistic
sources of information for target words embedded in sentential contexts. Eye-movement
measures for 43 native English speakers showed that a given contribution to representation (e.g.,
7
emotional embodiment) facilitated reading only when other contributions to representation (e.g.,
sensorimotor or linguistic) were insufficient for word recognition. Moreover, embodied effects
on word recognition were modulated by individual differences that attenuate and amplify
emotional and sensorimotor information, respectively. The results suggest that behavior is
functionally modulated by embodied information (i.e., emotional and sensorimotor) when
linguistic contributions to representation are not enhanced by high frequency. Furthermore,
emotional advantages are maximal when words are not already embodied by sensorimotor
contributions to representation (and vice versa). This work is consistent with recent studies that
suggest that abstract words are grounded in emotional experiences, analogous to how concrete
words are grounded in sensorimotor experiences.
The second study (Chapter 3) tests the hypothesis that word representations are
emotionally impoverished in a second language (L2), which has only been tested using tasks that
present words in isolation, or that require laboratory-specific decisions. In this study, eye
movements were recorded for 34 bilinguals who read sentences in their L2 with no goal other
than comprehension, and compared them to the 43 first language readers taken from Chapter 2.
Positive words were read more quickly than neutral words in the L2 across first-pass reading
time measures. However, this emotional advantage was absent for negative words for the
earliest measures. Moreover, negative words but not positive words were influenced by
concreteness, frequency, and L2 proficiency in a manner similar to neutral words. Taken
together, the findings suggest that only negative words are at risk of emotional disembodiment
during L2 reading, perhaps because a positivity bias in L2 experiences ensures that positive
words are emotionally grounded.
The third study (Chapter 4) investigates whether embodiment based on negative vs.
positive emotion have the same roles in representation and processing. Specifically, the study
tests whether negative and positive embodiment produce different effects during natural reading,
8
where at a given fixation people encode words that are currently fixated and to-be-fixated words
in the parafovea. Fifty native English speakers read emotional or neutral nouns in sentences.
Target nouns were preceded by definite articles, which were preceded by verbs (e.g., depicted the
angel). Parafoveal preview of nouns was valid or invalid before readers' eyes crossed a boundary
between the definite article and noun. The results showed that when people read the verb,
fixation durations for all first-pass measures were slower for valid nouns that were both
emotional and low frequent (when nouns were neutral, noun frequency did not influence reading
times on the verb). When people moved their eyes to the noun, however, gaze durations were
faster when there had been a valid preview of positive nouns that were high frequent. In
contrast, an emotional advantage for negative nouns was limited to a later stage processing
indexed by total reading time. Thus, for the sentences in this study, embodiment based on
negative emotion took longer to facilitate foveal processing of the noun than embodiment based
on positive emotion.
The findings in this dissertation show that word processing during sentence reading is
embodied by emotional and sensorimotor information. Thus, embodied experiences elicited by
words’ referents are activated for the purpose of word recognition. Moreover, these embodied
experiences are activated in a purely linguistic task that is ecologically valid, as the reading task
employed in this dissertation did not impose any goals on participants other than comprehension.
However, the findings also show that linguistic information can occasionally attenuate embodied
modulations of word recognition. Thus, linguistic information can sometimes be sufficient for
word recognition, leaving insufficient time for embodiment to functionally modulate word
processing.
9
ABRÉGÉ
Selon la conception classique des sciences cognitives, la signification des mots est
représentée par des symboles abstraits, intégrés à des modules cognitifs, qui ne peuvent accéder
directement aux informations associées aux modalités sensorimotrices. Cette conception ne cesse
d’influencer le développement théorique. Par exemple, les modèles formels de la reconnaissance
de mots (p. ex., Reichle, Rayner, & Pollatsek, 2003) omettent généralement d’inclure des
variables sémantiques qui considèrent ou prennent en considération l’expérience sensorimotrice
(voir aussi, Kuperman, Estes, Brysbaert, & Warriner, 2014). Depuis une vingtaine d’années, cette
conception est confrontée à une approche théorique alternative, qui propose plutôt que la
signification des mots est directement représentée au niveau des systèmes sensoriels et moteurs.
Cette théorie alternative appuie ainsi une vision embodied (c.-à-d.., incarnée) de la cognition
pour laquelle la signification des mots est « ancrée » dans la connaissance expérientielle (en
opposition aux symboles abstraits, tel que préconisé par la conception classique).
L’un des obstacles théoriques importants relatifs aux théories embodied de la cognition
concerne la difficulté d’ancrer des concepts abstraits et intangibles dont on ne peut directement
faire l’expérience. Par exemple, le mot « justice » possède un référent abstrait qui ne peut être
expérimenté directement. Pour remédier à cette difficulté, Vigliocco, Meteyard, Andrews, and
Kousta (2009) proposent que les mots possédants un référant concret et tangible (p. ex.,
« table ») sont ancrés dans les expériences sensorimotrices, alors que les mots possédants un
référant abstrait et intangible (p. ex., « justice ») sont plutôt ancrés dans les expériences
émotionnelles, en plus d’être représentés par l’information linguistique (p. ex., les associations
de mots). Suivant cette proposition théorique, la présente dissertation vise à évaluer la notion
d’embodiment émotionnelle dans la représentation des mots.
La première étude (Chapitre 2) vérifie si la contribution des émotions au traitement des
mots dépend uniquement des dimensions émotionnelles qui y sont associées. Se référant aux
10
approches embodied de la représentation de mot, cette étude propose ainsi d’examiner les
interactions entre les sources d’informations émotionnelles, sensorimotrices et linguistiques dans
le traitement de mots-cibles présentés dans le contexte de phrases. Les mesures de mouvements
oculaires de 43 individus, de langue maternelle anglaise, montrent que la contribution à la
représentation des mots (p. ex. l’embodiment émotionnelle) facilite la lecture seulement lorsque
les autres types de contributions à la représentation des mots sont insuffisants pour la
reconnaissance de mots. Les résultats montrent également que les effets embodied sur la
reconnaissance de mots sont modulés en fonction de différences individuelles qui atténuent les
informations émotionnelles et amplifient les informations sensorimotrices. Ces résultats
suggèrent que les informations dites embodied (c.-à-d.., émotionnelle et sensorimotrice)
modulent le comportement de manière fonctionnelle lorsque la contribution des informations
linguistiques aux représentations n’est pas améliorée par un niveau élevé de fréquence de mots.
De même, les bénéfices cognitifs liés aux dimensions émotionnelles du mot sont maximisés
lorsque celui-ci n’est pas déjà embodied par les contributions sensorimotrices aux représentations
de mot (et vice-versa). Dans l’ensemble, les résultats de cette étude sont cohérents avec d’autres
travaux récents qui suggèrent que les mots abstraits sont ancrés dans l’expérience émotionnelle
de façon analogue à la manière dont les mots concrets sont ancrés dans l’expérience
sensorimotrice.
La deuxième étude (Chapitre 3) évalue l’hypothèse selon laquelle les dimensions
émotionnelles des représentations de mot sont appauvries dans un contexte de langue seconde
(L2), une hypothèse qui a uniquement été testée dans des tâches expérimentales où les mots sont
présentés de manière isolée, ou dans des tâches qui requièrent une décision spécifique liée au
contexte expérimental. Dans cette étude, nous avons enregistré le mouvement oculaire de 34
individus bilingues lisant des phrases dans leur langue seconde (L2). Nous avons comparé les
données issues de cette étude, où l’unique objectif des participants était la compréhension des
11
mots, à celles collectées auprès des 43 individus de langue maternelle anglaise de l’étude
détaillée au chapitre 2. Cette comparaison indique que, dans la langue seconde, les mots à
connotation positive sont lus plus rapidement que les mots neutres, tel que mesuré lors du
premier passage de lecture. Ce bénéfice associé aux dimensions émotionnelles du mot était
néanmoins absent pour les mots à connotation négative lors des mesures précédentes. Il est aussi
intéressant de noter que différents aspects tels que le niveau de concrétude et la fréquence du
mot, ainsi que les habiletés individuelles dans la seconde langue influencent le traitement des
mots à connotation négative de façon similaire aux mots neutres, ce qui n’est pas le cas pour les
mots à connotation positive. Ensemble, ces résultats suggèrent que seuls les mots à connotation
négative sont à risque d’être dé-embodied (c.-à-d.., désincarnés) lorsqu’ils sont lus dans la langue
seconde. Cela suggère la présence possible d’un biais positif au niveau de la langue seconde qui
assurait l’embodiment émotionnel des mots à connotations positifs.
La troisième étude (Chapitre 4) évalue si l’embodiment basé sur le contraste entre les
émotions positives et négatives joue un rôle similaire au niveau des représentations de mot et du
traitement de l’information. Spécifiquement, cette étude teste si les types d’embodiment positif et
négatif produisent des effets différents dans un contexte naturel de lecture. Dans ce contexte, les
mots sont encodés tant au niveau du point de fixation du regard, qu’au niveau para-fovéal. Pour
cette étude, cinquante participants de langue maternelle anglaise ont lu des noms communs
émotionnellement chargés ou neutres, au sein de phrases. Les mots-cibles étaient précédés d’un
article défini et d’un verbe (p. ex., représentait l’ange). L’aperçu para-fovéal du mot à venir était
valide ou invalide au moment où l’œil du lecteur croisait un seuil limite invisible situé entre
l’article défini et le nom. Les résultats montrent que lorsque les participants lisaient le verbe, la
durée de fixation du regard durant le premier passage de lecture était plus lent pour les noms
valides émotionnellement chargés et peu fréquents (lorsque les noms étaient neutres, la fréquence
du mot n’a pas influencé le temps de lecture). Lorsque les participants dirigeaient leurs yeux vers
12
la cible, la durée du regard était moindre (ou plus rapide), lorsqu’il y avait eu, au préalable, un
aperçu valide des noms à connotation positive de fréquence élevée. En revanche, les résultats
indiquent que l’apport des dimensions émotionnelles des mots à connotation négative se limite à
un stade avancé du traitement de l’information, tel que mesuré par le temps de lecture total. Par
conséquent, en ce qui a trait aux phrases employées dans cette étude, l’embodiment basé sur
l’émotion négative a mis plus de temps à faciliter le traitement des noms au niveau fovéal que
l’embodiment basé sur l’émotion positive.
Les résultats dans cette dissertation confirment que le traitement des mots durant la
lecture est embodied par l’intermédiaire d’informations de type émotionnelle et sensorimotrice.
L’expérience dite embodied, suscitée par le référent du mot, est donc activée pour appuyer la
reconnaissance de mot. De plus, ces activations embodied proviennent de tâches purement
linguistiques et écologiquement valides, alors que la simple compréhension représentait la seule
exigence imposée aux participants durant les tâches de lectures rapportées dans cette dissertation.
Néanmoins, les résultats montrent aussi que l’information linguistique peut occasionnellement
atténuer la modulation embodied dans la reconnaissance de mot. Par conséquent, l’information
linguistique peut parfois être suffisante pour la reconnaissance de mot, ne laissant pas
suffisamment de temps à l’embodiment pour moduler fonctionnellement le traitement du mot.
13
FUNDING
Financial support for the research presented in this dissertation was provided by the
following support to Professor Debra Titone: A Natural Sciences and Engineering Research
Council of Canada (NSERC) Discovery Award, The Canada Research Chairs (CRC) Program,
Canadian Foundation for Innovation (CFI), Social Sciences and Humanities Research Council,
and the Center for Research on Brain, Language and Music (CRBLM). Additional financial
support was provided to Naveed Sheikh: An NSERC master’s scholarship, a fonds québécois de
la recherche sur la société et la culture master’s scholarship, and a Fond Québécois de recherché
sur la nature et les technologies doctoral scholarship.
14
STATEMENT OF ORIGINALITY
The research presented in this dissertation makes novel contributions to research on the
emotional embodiment of words. For example, each study in this dissertation investigates
whether embodiment effects on word processing generalize to natural reading using ecologically
valid eye-movement sentence reading tasks. This is a novel contribution to research on
embodiment in word representation because previous work and theories about embodiment are
based almost entirely on data collected with single-word paradigms. In single-word paradigms,
words are generally presented in isolation without a context and/or require participants to make
laboratory-specific responses. Thus, the research presented in this dissertation shows that
embodied theories generalize to natural language processing.
Each individual study presented in this dissertation also makes unique contributions to
research on embodiment in word representation. The first study presented in this dissertation
(Chapter 2), published in Emotion, investigates how native speakers of English process
emotional target words while reading sentences in their first language (L1). The novel
contribution of this study is that it demonstrates how emotional contributions to word
representation modulate word processing as a function of sensorimotor and linguistic
contributions to representation. Moreover, this study investigates whether embodiment effects
are modulated by individual differences in sensitivity to emotional and sensorimotor information,
which are also novel contributions to research on embodiment.
The second study presented in this dissertation (Chapter 3), published in Cognition and
Emotion, investigates how French-English bilinguals (native speakers of French) process
emotional target words while reading sentences in their second language (L2) (English). The
novel contribution of this study is that it demonstrates that embodiment based on negative vs.
positive emotion have different representational roles in bilinguals processing words in their L2.
Moreover, this study investigates how individual differences in L2 proficiency among bilinguals
15
predict embodiment based on different sources of experiential knowledge, which is also a novel
contribution to research on embodiment.
The third study presented in this dissertation (Chapter 4), submitted for publication to
Emotion, investigates how native speakers of English process emotional target words while
reading sentences in their L1. The novel contribution of this study is that it investigates how
emotional target words parafoveally influence word processing, before they are directly fixated.
Moreover, this study demonstrates that embodiment based on negative vs. positive emotion
produce different foveal vs. parafoveal word processing effects, which is also a novel
contribution to research on embodiment.
16
CONTRIBUTION OF AUTHORS
Chapter 2:
Sheikh, N. A., & Titone, D. A. (2013). Sensorimotor and linguistic information attenuate
emotional word processing benefits: An eye-movement study. Emotion, 13, 1107–1121.
doi:10.1037/a0032417
As first author, I was responsible for the entire research process. This included the
conception and design of the research and doing the background research. I was also responsible
for creation of the stimuli, the programming of the experiment, and managing the data collection
process. I was also responsible for data analysis and interpretation, and writing the majority of
the manuscript
As senior author, Dr. Debra Titone contributed to and supervised all aspects of the
research process. This included the conception and design of the study, the preparation of the
stimuli, interpretation of the data analyses, and writing and revising the manuscript.
Chapter 3:
Sheikh, N. A., & Titone, D. (2015a). The embodiment of emotional words in a second language:
An eye-movement study. Cognition and Emotion.
As first author, I was again responsible for the entire research process, including the
conception and design of the research, and the doing the background research. I was also
responsible for creation of the stimuli, the programming of the experiment, and managing the
data collection process. I was also responsible for data analysis and interpretation, and writing
the majority of the manuscript
As senior author, Dr. Debra Titone contributed to and supervised all aspects of the
research process. This included the conception and design of the study, the preparation of the
stimuli, interpretation of the data analyses, and writing and revising the manuscript.
17
Chapter 4:
Sheikh, N. A., & Titone, D. A. (2015b). The parafoveal and foveal effects of negative and
positive emotional embodiment: Evidence from eye movements and the gaze-contingent
boundary paradigm. Manuscript submitted for publication to Emotion.
As first author, I was again responsible for the entire research process, including the
conception and design of the research, and doing the background research. I was also
responsible for creation of the stimuli, the programming of the experiment, and managing the
data collection process. I was also responsible for data analysis and interpretation, and writing
the majority of the manuscript
As senior author, Dr. Debra Titone contributed to and supervised all aspects of the
research process. This included the conception and design of the study, the preparation of the
stimuli, interpretation of the data analyses, and writing and revising the manuscript.
18
CHAPTER 1:
GENERAL INTRODUCTION
19
GENERAL INTRODUCTION
Language comprehension is possible only if people share basic ways of perceiving the
world (Tomasello, 2003). However, a hotly debated question within cognitive science is whether
perceptual experiences directly play a role in language comprehension (along with the myriad
other sensations that people experience). For example, consider a delicious steak. People can
see, smell, and taste steak; physically manipulate it in any number of ways (e.g., grill, sear, sousvide, etc.), and have preferences about its doneness (e.g., blue rare, rare, medium rare, etc.).
According to traditional views of cognition, online language comprehension does not depend on
systems that represent these experiences. Instead, the information within these experiences is
transduced into an abstract and self-contained language of thought or “mentalese,” and online
language comprehension is based on activations within this self-contained system (Fodor, 1975,
1983; see also Pylyshyn, 1999). Thus, according to the traditional view, people understand the
meanings of words like steak without having to activate systems that represent experiences
associated with eating steaks. However, the traditional view is being challenged by an
alternative theoretical perspective, generally called an embodied approach to cognition, which
proposes that knowledge is represented by (i.e., grounded in) the brain’s systems for interacting
with the world (Barsalou, 1999; Glenberg, 1997; Lakoff & Johnson, 1980; Wilson, 2002; Zwaan,
2008). In embodied approaches to cognition, the conceptual representation for steak is thought
to be grounded in the brain’s systems for perception, action, and introspection (Barsalou, 1999).
In other words, perception and conceptualization are based on the same representations
according to embodied approaches to cognition. Thus, this debate concerns age-old questions
about the very nature of thought. As I will discuss below, the traditional view has dominated
psychology since the cognitive revolution. However, embodied approaches to cognition have
now been extended to virtually all traditional areas of psychology and beyond, including the
issue of how people extract meaning from language (Glenberg, 2010).
20
The core assumption of the traditional view in cognitive science is that the brain's
systems for experiencing the world are encapsulated and separated from cognition (e.g., Fodor,
1983). This traditional view can be characterized in terms of a computer: Cognition is a central
information processor, and systems for interacting with the world are peripheral slave systems
that plug into central cognition. Thus, central cognitive mechanisms must operate on abstract
amodal symbols because modality-specific information is encapsulated within the slave systems.
However, in embodied approaches to cognition, cognitive processes operate on modality-specific
information. For example, in Barsalou’s (1999) embodied approach to cognition, called
perceptual symbol systems, cognitive mechanisms operate on modal symbols produced by
perceptual states. Similarly, Lakoff and Johnson (1980) and Glenberg (1997) proposed that the
content in perceptual information was mapped into abstract ideas expressed in language. Thus,
embodied theories propose that cognitive mechanisms have direct access to perceptual
information. The computer metaphor cannot be straightforwardly extended to incorporate a
comparable mechanism because it assumes that perceptual information is locked away in slave
systems, which are impenetrable to central cognitive processes (Pylyshyn, 1999). Although the
computer metaphor has traditionally dominated cognitive science, embodied approaches have
been gaining ground over the last 15 years, particularly with respect to language comprehension
(Zwaan, 2014), and could even serve as a unifying principle for all of psychology (Glenberg,
2010).
In spite of the uptick in research that has been driven by embodied theories, embodied
approaches to word representation have not been applied to several open questions in the word
processing literature. These gaps are also boundaries that undermine an embodied account of
word representation. For example, embodied theories of word representation have focused
primarily on concrete words (e.g., table) and perceptual/motor information—it is not clear to
what extent embodied ideas generalize to abstract words (e.g., justice) and emotional
21
information, which play fundamental roles in human experience. In addition, embodied theories
have focused on native speakers processing words in their first language (L1)—it is not clear to
what extent embodied ideas generalize to the situation of when people who know more than one
language (i.e., bilinguals) process words in their second language (L2). Such questions about
bilingualism are important given that people who speak more than one language make up a large
proportion of the global population (Grosjean, 2010). Finally, embodied theories have been
tested primarily with laboratory-specific tasks in which individuals respond to words that are
presented in isolation—it is not clear to what extent embodied ideas generalize to ecologically
valid tasks like eye-movement sentence reading paradigms, in which individuals read words as
part of a sentence with no goal other than comprehension. The goal in this thesis is to push each
of these boundaries that currently limit our understanding of embodiment in natural language
processing.
In what follows, I first review embodied approaches to word representation, with
particular emphasis on the role of emotion. I follow this introduction with a review of key
studies on L1 and L2 emotional word processing, and how they may connect with an embodied
perspective. Finally, I present background information on the field of eye movements during
reading, given the relevance of this method to the current dissertation.
Embodied Word Representations
An embodied approach to word representation is consistent with a perceptually based
view of the mind that persisted, until very recently, for 2,000 years (Barsalou, 1999). For
example, Barsalou argues that cognition has been viewed as imagistic since Aristotle and
Epicurus in 4th century BC. Similar thinking underlay the approach of Wilhelm Wundt, who
established the first psychological research laboratory at the University of Leipzig in 1879
(Boring, 1950). Specifically, in one line of work, Wundt examined how sensation was
compounded into ideas, which he believed were not fundamentally different from perceptual
22
experiences (Carpenter, 2005; Thomas, 2014; Wundt, 1897). This perceptually based view of the
mind went out of style under behaviorism, which made the study of imagery, and mental states
more generally, verboten. This attitude persisted even after the cognitive revolution:
Psychologists were reticent to study imagery, for fear that it was viewed as too subjective to
study empirically (Benjafield, Smilek, & Kingstone, 2010). It is not clear what underlay this
reticence. However, Barsalou (1999) proposed that there could have been lingering paranoia
regarding the attacks of behaviorists after the cognitive revolution.
The dominant view in psychology became that perceptual information is transduced out
of conceptual representations, making them amodal symbols (i.e., modality-specific information
is stripped away from them). This amodal view is consistent with modular views of the mind,
and the notion of modularity is deeply entwined with views on language. For example,
Chomsky proposed that a language acquisition device (LAD), a hypothetical module of the
mind, explains how children acquire language with insufficient linguistic input from the
environment (see Chomsky 1959, 1968, 1980, 1986, for Chomsky’s arguments that the
environment does not provide sufficient linguistic input for language acquisition). Although the
theoretical need for an LAD has been questioned more recently (e.g., Tomasello, 2003), the LAD
was highly influential at the time. Fodor (1983) drew on this idea to formulate his own views on
modularity, which were spelled out in his book, Modularity of Mind. The book re-popularized
the modularity idea, which had withered potentially because of its association with movements in
psychology, such as phrenology (Fodor, 1983). For the phrenologists Franz Joseph Gall (17581828) and his student J.G. Spurzheim (1776-1832), modularity went hand in hand with
localization, as they thought that highly specialized modules were housed in very specific
regions of the brain (Benjafield et al., 2010). Moreover, they infamously believed that
measurable protrusions on the skull provided insight into how developed or underdeveloped
particular modules may be, which had a transformative influence on 19th century psychological
23
counselling and phrenologists often provided advice to paying clients (Sokal, 2001). The
diagnostic value of skull protrusions were no longer taken seriously by the 1840s (McGrew,
1985). However, the phrenologists’ ideas about localization of function continue to receive
attention. For example, around the same time that Fodor published Modularity of Mind, Gardner
(1983) pointed to phrenology as a lineal ancestor to his work on multiple intelligences housed in
the brain. In Fodor's modularity thesis, the focus was on properties that modules must
implement “to some interesting extent” (Fodor, 1983, p. 37). However, the most important
aspect of modularity in his thesis was what he called informational encapsulation. This
encapsulation renders information within a module cognitively impenetrable to processes outside
that module; for example, encapsulation would make visual perception impenetrable to cognition
(see Pylyshyn, 1999).
While modular views of the mind dominated the theoretical landscape, Paivio (1965;
1969; 1971) managed to renew interest in imagery, and his ideas were similar to the notion of
embodiment. Paivio’s success may be due in part to his straightforward operationalization: He
operationalized imagery as the ease with which words elicit a mental image. Paivio eventually
proposed dual-coding theory, which accounts for differences among words in mental imagery
(Paivio, 1971). The theory accounts for imagery differences in terms of representational codes
that encode two types of information. One type of information, which underlies word use, is
encoded by verbal codes, called logogens, in a verbal system. Another type of information,
which generates mental images, is encoded by imagistic codes, called imagens, in a nonverbal
system. This theory accounts for processing differences between concrete and abstract words,
which have different types of referents. For example, the referents of concrete words like chair
can be directly experienced by the senses. In contrast, the referents of abstract words like idea
are less directly experienced by the senses. Several studies show that responses are faster for
concrete words compared to abstract words in word processing tasks like lexical decision (e.g.,
24
Binder, Westbury, McKiernan, Possing, & Medler, 2005; James, 1975; Kroll & Merves, 1986),
and that concrete words have an advantage over abstract words in memory tasks (e.g., Allen &
Hulme, 2006; Paivio, 1971; Romani, McAlpine, & Martin, 2007). This advantage for concrete
words is called the concreteness effect. Dual-coding theory accounts for the concreteness effect
by assuming that concrete words have richer representations than abstract words. Specifically,
the theory assumes that concrete words are represented in both the verbal and imagistic systems,
because the experience of concrete events is saturated with images. In contrast, abstract words
are represented primarily or exclusively in the verbal system. Thus, the concreteness effect is
attributed to concrete words having representation in the verbal and imagistic systems, whereas
abstract words generally have representation in only the verbal system.
Paivio did not explicitly formulate dual-coding theory as an embodied theory. However,
the claim in dual-coding theory that an activity of the mind (i.e., processing concrete words) is
based on a system that evolved for interacting with the environment (i.e., visual imagery) makes
it an embodied approach to word representation, at least where concrete words are concerned,
because that is a claim made by embodied theories of cognition (Wilson, 2002). In contrast,
dual-coding theory is not embodied where abstract word processing is concerned. Instead, the
theory proposes that abstract word processing relies on verbal codes in a verbal system, and the
theory does not make clear whether these verbal codes represent words’ “linguistic form or
linguistic meaning” (Vigliocco, Meteyard, Andrews, & Kousta, 2009, p. 231).
This omission touches on an often cited problem in embodied theories: How does a
theory that grounds knowledge in experiential information account for abstract words, which
have referents that cannot be directly experienced (Barsalou, 1999; Lakoff & Johnson, 1980;
Glenberg, 1997)? One approach is to assume that amodal linguistic knowledge is sufficient for
representing abstract words, as dual-coding theory implies. Abstract words would then be
represented in terms of linguistic information (e.g., relationships with other words), leaving
25
abstract semantics disembodied. However, Glenberg and Robertson (2000) have argued that
abstract symbols like English words must be grounded in something other than more abstract
symbols to have meaning. Harnad (1990) also addressed this topic in his paper on the symbol
grounding problem. To elucidate this problem, he describes the task of learning Chinese as an
L1 using only a Chinese/Chinese dictionary (i.e., similar in some respects to Searle’s classic
Chinese Room thought experiment; Searle, 1980). Although a learner could learn to differentiate
and recognize different words on the basis of the visual patterns of the linguistic symbols, it
would be impossible to acquire the meanings of words without referring to experiences outside
the language. This is the symbol grounding problem, which a disembodied approach to abstract
semantics would suffer, as would any traditional amodal approach to semantics more generally.
A potential embodied approach to this problem, originating in cognitive linguistics,
assumes that abstract semantics are metaphorically grounded in the same systems as concrete
semantics (Lakoff & Johnson, 1980). In their landmark publication, Metaphors We Live By
(1980), Lakoff and Johnson argued that metaphors structure cognition. For example, argument
can be understood in terms of war (e.g., claims can be indefensible, weak points can be attacked,
etc.). Similarly, anger can be understood in terms of a liquid exploding out of a container, due to
rising pressure. In another landmark publication on embodied cognition, Glenberg (1997) also
addressed how people might understand abstract language. Glenberg (1997) cited arguments
made by Talmy (1988) on how abstract causal terms (e.g., because) can be understood in terms
of one entity exerting some force on another. As an example, Glenberg (1997) described how the
sentence, “The ball kept rolling because of the wind blowing on it,” can be interpreted in terms
of one entity tending toward inaction (i.e., the ball) being acted upon by another entity tending
toward action (i.e., the wind). Thus, both Glenberg (1997) and Lakoff and Johnson (1980)
proposed that conceptualizing abstract situations in concrete ways is sufficient for
comprehension. In other words, they suggested that the concrete domain can be mapped onto an
26
abstract domain to metaphorically ground abstract semantics in concrete events (Barsalou, 1999).
However, there are problems with this approach as well. For example, recall the example of
anger being understood in terms of a liquid exploding from a container. Barsalou (1999) argued
that an adequate understanding of anger can hardly be achieved only by knowing that anger is
like liquid exploding from a container. Barsalou also pointed out that it would not be possible to
systematically map concrete knowledge into an abstract domain that has no content. Thus, it
remains essential to specify a direct non-metaphorical representation of the abstract domain.
Direct Representation in Sensorimotor Vs. Emotional Experience
These arguments regarding the content of the abstract domain suggest that the key
objective in building a complete theory of word representation, one that accounts for concrete
and abstract semantics, is to identify the type of information that directly and non-metaphorically
represents the abstract domain. A promising new theory proposes that emotion may serve as the
foundational content that directly represents abstract semantics (Vigliocco et al., 2009).
Vigliocco et al. (2009) posited that there are two sources of experiential information that ground
word meaning. One source of experiential information consists of sensory events (e.g., vision)
and actions (e.g., movements) in the external world. This is the source of information that has
received the most attention in embodied theories of cognition. The second source of experiential
information consists of emotional states in the inner world (e.g., negative and positive emotion).
For example, consider the words apple and grief. People have very different bodily experiences
with the referents of these words. For example, apples are tangible objects that people can grasp,
eat, or physically manipulate in any number of ways (e.g., slice, stew, puree etc.), which are
generally classed as sensorimotor experiences. In contrast, grief is a less tangible concept that
people may experience through indirect physical manifestations (e.g., a lump in one’s throat),
which are generally classed as emotional experiences. Vigliocco et al. (2009) proposed that
concrete words generally co-occur more often with sensorimotor experiences, and that abstract
27
words generally co-occur more often with emotional experiences. They also proposed that
concrete and abstract semantics are primarily grounded in these sensorimotor and emotional
sources of information, respectively, consistent with the idea that words are grounded in the
experiences that they tend to co-occur with during language use (Zwaan, 2008). In addition to
these embodied sources of information, Vigliocco et al. (2009) posited that knowledge is also
represented by purely linguistic sources of information (e.g., verbal associations and syntactic
information) (see also Barsalou, Santos, Simmons, & Wilson, 2008; Louwerse & Jeuniaux, 2008;
Louwerse & Jeuniaux, 2010; Zwaan, 2008).
One approach to test the proposal in Vigliocco et al. (2009) is to instruct participants to
attend to the meaning of abstract words, in order to directly examine their representations. For
example, people can be asked to rate the concreteness of words, and emotional words should, on
average, be rated as more abstract if emotional experiences co-occur more often with abstract
words than concrete words. Or conversely, abstract words should be rated as more emotional.
Vigliocco and colleagues (Kousta, Vigliocco, Vinson, Andrews, & Del Campo, 2011) cited
research by Altarriba, Bauer, and Benvenuto (1999), who reported exactly this finding. Altarriba
et al. had participants rate the concreteness of concrete words, abstract words, and emotion
words that directly referred to emotional states (e.g., anger). They found that the most abstract
words more likely refer directly to emotional states. Vigliocco and colleagues (Kousta et al.,
2011) also cited research by Barsalou and Wiemer-Hastings (2005), who reported findings which
suggest that inner states may ground abstract semantics. In their study, Barsalou and WiemerHastings asked participants to generate features for concrete and abstract words. Their results
showed that abstract words had more features that focused on introspective content compared to
concrete words, consistent with the idea that abstract words may be grounded in emotional states.
Vigliocco and colleagues also examined whether emotional words tend to be more abstract in
Vigliocco et al. (2013). In this study, they examined whether concreteness ratings are predicted
28
by emotionality. Emotionality was measured in a norming study along two dimensions, valence
and arousal, consistent with non-discrete approaches to emotion (Russell, 2003; Russell &
Barrett, 1999). Valence reflects the extent to which a word is emotionally charged (i.e.,
emotionally negative or positive, which are reflected by low and high valence, respectively;
emotional neutrality is reflected by moderate valence ratings), and arousal reflects the extent to
which a word is calming or exciting (i.e., low and high arousal, respectively). Consistent with
the idea that abstract words are more emotional than concrete words (Vigliocco, 2009), Vigliocco
et al. (2013) found that emotionally negative and positive words were more abstract compared to
words that were emotionally neutral. They also found a similar effect for arousal: Words with
high arousal ratings were more abstract compared to less arousing words.
Another approach to test the proposal in Vigliocco et al. (2009) is to examine the
processing effects of emotion and concreteness without explicitly asking participants to attend to
word meaning. Vigliocco and colleagues took this approach in Kousta et al. (2011). In this
study, Kousta et al. (2011) performed large-scale regression analyses using lexical decision data
from the English Lexicon Project (Balota et al., 2007). Surprisingly, they found that lexical
decisions were faster for abstract words compared to concrete words, which is a reversal of the
typical concreteness effect (where the advantage is for concrete words). They called this reversal
the abstractness effect, and hypothesized that the reversal occurred because the abstract words in
their data set were more emotionally charged compared to the concrete words. Vigliocco and
colleagues had previously found faster lexical decisions for emotional words compared to neutral
words in a previous study that was based on the same database of stimuli (Kousta, Vinson, &
Vigliocco, 2009). Thus, a valence-based hypothesis provided a straightforward explanation for
the surprising abstractness effect in Kousta et al. (2011), which they confirmed with follow-up
analyses in the same study. Specifically, they found that there was no abstractness effect in
subsets of words in which valence and concreteness were unconfounded, or if valence was
29
statistically controlled. Kousta et al. (2011) concluded that their results provide additional
support for the idea in Vigliocco et al. (2009) that abstract words are grounded in emotional
experiences.
However, one open question concerns whether the three representational sources of
information identified in Vigliocco et al. (2009), emotional, sensorimotor, and linguistic,
simultaneously modulate word processing. For example, how does emotionality affect the
processing of words that vary in their sensorimotor contribution (i.e., concreteness), when
emotionality is not confounded with concreteness? Moreover, how do emotionality and
concreteness affect the processing of words that vary in their linguistic contribution?
One potential mechanism for understanding how embodied and linguistic information
might jointly shape word processing has been described within a feedback activation framework
(Newcombe et al., 2012; see also Hino & Lupker, 1996; Pexman, Hargreaves, Edwards, Henry,
& Goodyear, 2007; Pexman, Lupker, & Hino, 2002; Siakaluk, Pexman, Aguilera, Owen, &
Sears, 2008; Siakaluk, Pexman, Sears, Wilson, Locheed, & Owen, 2008; Yap, Tan, Pexman, &
Hargreaves, 2011; Zdrazilova & Pexman, 2013). In this framework, when people encounter
words, they activate memory representations of those words across several levels (e.g., their
visual characteristics or orthography, their meaning characteristics or semantics). Following
initial bottom-up activation across these levels, there is rapid feedback of activation from higher
semantic levels to lower orthographic levels of representation, which enables the meanings of
words to directly affect how words are perceived and recognized. The framework assumes
greater feedback when words have semantic representations that are rich, either by virtue of
having many associated semantic features or semantic features that are enriched by embodied
experiences (Newcombe et al., 2012). Thus, this framework predicts that embodied semantics
can influence performance even in tasks where responses can, in principle, be made purely on
the basis of linguistic information (e.g., deciding whether letter strings refer to words or
30
nonwords in a particular language, as in a lexical decision task).
Of note, the semantic representations grounded in emotional and sensorimotor
experiences are presumably distributed rather than localist in nature (i.e., there is no
circumscribed place in the brain that corresponds one-to-one with a word’s meaning; rather,
different aspects of a word's meaning are distributed across different areas of the brain and likely
overlap with other words’ meanings). Thus, they may be neurally implemented in a manner
similar to how action words have somatotopic representation in sensorimotor regions of the
brain, in addition to regions traditionally associated with linguistic representation (Hauk,
Johnsrude, & Pulvermüller, 2004). Although these distributed networks have not yet been fully
realized for abstract and concrete words (Kousta et al., 2011), neuroimaging work in the
semantic feedback literature shows that words associated with richer semantics settle more
quickly than words associated with less semantic information (Pexman et al., 2007). Thus, the
semantic feedback framework predicts that rich embodied semantics based on emotionality
should produce faster processing for emotional compared to neutral words, and that rich
embodied semantics based on concreteness should produce faster processing for concrete
compared to abstract words (i.e., when emotionality and concreteness are not confounded; see
Kousta et al., 2011).
However, feedback from embodied semantics may not always influence responses in
word processing tasks. This topic is the focus of a different theory called the linguistic and
situated simulation theory (Barsalou, Santos, Simmons, & Wilson, 2008). The linguistic and
situated simulation theory describes how embodied and linguistic information in word
representations jointly shape responses, and specifically addresses the circumstances when
embodied information will not functionally modulate responses. The theory's predictions are
based on three assumptions. First, the theory assumes that embodied and linguistic information
are immediately activated upon perception of a word. Second, it assumes that the activation of
31
linguistic information peaks earlier than embodied information because linguistic information
(e.g., orthographic information) is more similar to the perceived stimulus word (see also Tulving
& Thomson, 1973). Finally, linguistic information is thought to represent relatively superficial
information like words' orthographic forms or associations between words, in contrast to
embodied information which represents experiences that ground the semantic meanings of
words. On the basis of these assumptions, the linguistic and situated simulation theory proposes
that embodied semantic information functionally modulates responses only if linguistic
information is insufficient for a response. As an example, Barsalou et al. (2008) discuss how
embodied and linguistic information influence the decision that a letter string refers to a word or
a nonword in a particular language (i.e., a lexical decision task). The authors propose that when
nonwords consist of letter strings that are orthographically and phonologically illegal (i.e., they
cannot be real words in the language), linguistic information will be sufficient for quickly
reaching a decision and generating a response. Thus, there would likely be insufficient time for
embodied semantics to functionally modulate responses. However, if the nonwords consist of
letter strings that are orthographically and phonologically legal (i.e., they look like they could be
real words in the language), then linguistic information may not be sufficient for quickly
generating a response, leaving more time for the activation of embodied semantics to
functionally modulate responses. In other words, emotional and sensorimotor information
should not modulate word processing when linguistic information is sufficient for a response.
Research from Juhasz, Yap, Dicke, Taylor, and Gullick (2011) suggests that one approach
to test linguistic and situated simulation theory is to examine interactions between embodied
information (i.e., emotionality and concreteness) and word frequency. Word frequency refers to
the frequency with which words occur in natural language, which is typically estimated using
text corpora (e.g., The HAL corpus; Burgess & Livesay, 1998). In word processing tasks,
frequent words (e.g., table) are recognized more quickly than infrequent words (e.g., credenza)
32
(e.g., Balota & Chumbley, 1984; Monsell, Doyle, & Haggard, 1989; Rayner & Duffy, 1986),
which is called the frequency effect. Frequency effects are thought to reflect the repeated access
of linguistic information within and across languages (e.g., lexical forms; see Whitford & Titone,
2012; Gollan et al., 2011), and word frequency accounts for a large proportion of variance in
word processing (Balota et al., 2004; Brysbaert & New, 2009; Yap & Balota, 2009).
Interestingly, word frequency often modulates other factors that affect word processing (e.g.,
Cortese & Schock, 2013; Gerhand & Barry, 1999a; 1999b), and the results in Juhasz et al. (2011)
suggest that frequency may also modulate embodied effects on word processing. Juhasz et al.
examined the extent to which words evoke embodied experiences in a norming study, and these
words also varied in word frequency. They found that word frequency is negatively correlated
with the extent to which words evoke embodied experiences, suggesting that high-frequency
words are less likely to evoke embodied experiences compared to low-frequency words. Juhasz
et al. concluded that high-frequency words, compared to low-frequency words, might rely more
purely on linguistic forms of processing, consistent with the linguistic and situated simulation
theory (Barsalou et al., 2008), given the reasonable assumption that word frequency makes a
linguistic contribution to word representation.
An issue not directly addressed in the linguistic and situated simulation theory (Barsalou
et al., 2008) is whether the effect of an embodied source of information (e.g., concreteness) on
word processing can also be modulated by a different embodied source of information (e.g.,
emotionality). However, it is possible to derive a prediction that is consistent with the linguistic
and situated simulation theory. Specifically, the theory essentially predicts that any one
representational aspect will not functionally modulate processing if any other representational
aspect is sufficient for generating a response. In other words, any given embodied aspect of
representation, whatever its source, will maximally influence processing when other aspects of
representation, whether embodied or linguistic, are insufficient for generating a response. Thus,
33
a reasonable extrapolation of the linguistic and situated simulation theory is that emotionality
should maximally facilitate processing when words are abstract and low in frequency, and
concreteness should maximally facilitate processing when words are emotionally neutral and low
in frequency.
To summarize, recent work suggests that emotion plays a foundational role in grounding
semantic representations, analogous to the role played by sensorimotor information (Vigliocco et
al., 2009). Moreover, embodied theories suggest that emotional, sensorimotor, and linguistic
information should jointly influence processing (e.g., Barsalou et al., 2008). In the following
sections, I review relevant studies that examined emotionality and concreteness effects (though
most of these studies generally did not frame their results in terms of embodied cognition). In
addition, this review will help illuminate two important issues not addressed by the embodied
theories just reviewed. First, the review will make clear that embodiment based on negative vs.
positive emotion will, at least some of the time, exert different effects on word processing, which
is not addressed by the embodied framework described above. As I will discuss below, this
thesis proposes two potentially fruitful ways to contend with this issue: One based on individual
differences that may differentially modulate the effects of negative vs. positive valence (and
other types of embodiment more generally), and another based on an analysis of task demands
across studies that may differentially modulate the effects of negative vs. positive valence.
Second, the review will also make clear that bilinguals may not show the same emotional effects
in their L2 as native speakers in their L1, which also is not addressed by the embodied
framework described above. As I will discuss below, this thesis proposes that a consideration of
bilinguals' behavioral ecology may be a fruitful approach to understanding differences between
L1 and L2 emotional word processing.
Also of note, another issue that the review will make clear is that the vast majority of
studies on emotional word processing are based on tasks in which words are presented in
34
isolation and which require participants to produce laboratory-specific responses (e.g., lexical
decisions). This raises questions about the ecological validity of emotional effects on word
processing. Moreover, the ecological validity question also applies to the embodied approach
described above because it is also based on data produced by laboratory-specific responses. If
embodied approaches are to help build a complete theory of word representation, it is crucial to
demonstrate that embodied theory does not apply solely to experimental tasks in which
participants make laboratory-specific responses. Thus, each question raised in this thesis will be
addressed using an ecologically valid paradigm that combines a sentence processing task with
eye-movement measures of reading, which imposes no goal on participants other than
comprehension.
L1 Emotional Word Processing
Most studies on emotional word processing focus on native speakers processing words in
their L1 (e.g., Holt et al., 2006; Holt, Lynn, & Kuperberg, 2009; Kousta et al., 2009; Kousta et
al., 2011; Kuchinke, Võ, Hofmann, & Jacobs, 2007; Long & Titone, 2007, Méndez-Bértolo,
Pozo, & Hinojosa, 2011; Scott, O'Donnell, Leuthold, & Sereno, 2009; Scott, O’Donnell, &
Sereno, 2012). Early studies commonly employed the emotional Stroop task (e.g., McKenna &
Sharma, 1995; Pratto & John, 1991; Wentura, Rothermund, & Bak, 2000). This task requires
participants to name the font color of words while ignoring their meanings (as in the original
color Stroop paradigm, Stroop, 1935). The words in the emotional Stroop task spell out
emotionally charged or emotionally neutral meanings. The typical finding in these studies was
slower responses for negative words compared to neutral words (for a review, see Williams,
Mathews, & MacLeod, 1996). However, a meta-analysis of 32 such emotional Stroop studies
showed that emotional words were longer in length and/or lower in frequency than neutral words
(and/or had some other lexical confound), which could account for slower responses for negative
compared to neutral words (Larsen, Mercer, & Balota, 2006).
35
These methodological concerns have been addressed in subsequent studies. However, the
effect of valence remains unclear. For example, Estes and Adleman (2008) and Larsen, Mercer,
Balota, and Strube (2008) did large-scale regression analyses using lexical decision and naming
data from the English Lexicon Project (ELP; Balota et al., 2007), using stimulus words from the
Affective Norms for English Words (ANEW; Bradley & Lang, 1999). Estes and Adleman
(2008) found that responses were generally slower for negative words than positive words.
Unlike Estes and Adleman, Larsen et al. (2008) tested interactions between valence and arousal,
and found that only responses to negative words with particular arousal profiles were slower
compared to responses to positive words.
In contrast to these studies, Kousta et al. (2009) found no differences between negative
and positive valence in a lexical decision experiment in which they tested their own participants.
Instead, they found that responses were faster for negative and positive words compared to
neutral words. This group also replicated their finding using large-scale regression of ELP data
in the same study, and also found that the valence effect for negative and positive words did not
depend on arousal. They also replicated their finding again more recently (Vinson, Ponari, &
Vigliocco, 2014) using lexical decision data from the British Lexicon Project (Keuleers, Lacey,
Rastle, & Brysbaert, 2012).
Another subset of recent studies focused on interactions between emotional valence and
word frequency. For example, Kuchinke et al. (2007) and Scott et al. (2009) compared lexical
decisions for words varying in emotional valence and word frequency. For low-frequency items,
they found faster responses for negative and positive words compared to neutral words, and no
differences between negative and positive words. But for high-frequency items, only responses
to positive words were faster compared to neutral words. Recently, Kuperman, Estes, Brysbaert,
and Warriner (2014) also found interactions between emotional valence and word frequency in a
large-scale regression analysis of ELP data using the largest set of items to date. They found that
36
among low-frequency words, responses were faster for positive words compared to negative and
neutral words, and slower for negative words compared to neutral words. They also found that
the emotional effects for negative words and positive words vs. neutral words were attenuated
among higher frequency words. Moreover, as in Kousta et al. (2009), Kuperman et al. found that
these valence effects did not depend on arousal. Recently, Scott et al. (2012) used eyemovement measures of sentence reading to test interactions between emotional valence and word
frequency for target words embedded in emotionally neutral sentences. They found that people
read both negative and positive words faster than neutral words when those words were low in
frequency. But for high-frequency words, only positive words were read faster than neutral
words.
Although the findings vary, particularly with respect to the role of valence polarity, a
common pattern is faster responses for emotional words compared to neutral words. Moreover,
this emotional advantage is more often observed for low-frequency words. Interestingly, a small
number of lexical decision studies that tested interactions between concreteness and frequency
found a similar pattern: Faster responses for concrete words compared to abstract words is more
commonly observed among low-frequency items (James, 1975; Kroll & Merves, 1986; see also
de Groot, 1989). Thus, these findings support the idea that feedback from embodied semantics,
whether based on emotionality or concreteness, modulates early linguistic/lexical processing
(Newcombe et al., 2012). Moreover, this feedback from embodied semantics appears to
maximally influence responses when words are low in frequency (though there may be
exceptions for positive words). Thus, the findings are mostly consistent with the linguistic and
situated simulation theory (Barsalou et al., 2008).
However, multiple questions are left open by these data. For example, do emotional
valence, concreteness, and frequency jointly shape processing, consistent with an embodied
approach to word representation? Another question left open is why negative valence and
37
positive valence sometimes exert different effects on word processing? Both of these questions
will be investigated in this thesis.
L2 Emotional Word Processing
Although there remain unresolved questions regarding L1 emotional word processing, the
review in the preceding section shows that emotion plays an important role in word processing
for native speakers using their L1. However, emotionally charged words lead a double life
within the L2 literature. Some studies conclude that L2 words are emotionally impoverished
because differences between emotional words and neutral words are reduced during L2
processing compared to L1 processing (e.g., Degner, Doycheva, & Wentura, 2012). However,
other studies do not find L1-L2 differences (e.g., Sutton, Altarriba, Gianico, & Basnight-Brown,
2007), producing fractures in the empirical landscape that have not yet been explained. One
problem is that there are relatively few studies on L2 emotional word processing, and most of
these use different tasks that have different comprehension demands. Moreover, bilinguals might
also process emotional words differently as a function of L2 proficiency and emotional polarity.
Thus, another goal in this thesis is to test these alternatives and determine whether an embodied
theoretical approach to language can explain when bilinguals process L2 emotional words like
L1 emotional words.
One source of complexity regarding L2 emotional word processing is that studies vary
both in terms of experimental tasks and the L2 proficiency of bilinguals. For example, studies
that used the emotional Stroop task to test bilinguals that are as proficient in their L2 as their L1
(Eilola, Havelka, & Sharma, 2007) or more proficient in their L2 than their L1 (Sutton, Altarriba,
Gianico, & Basnight-Brown, 2007) found intact emotional word processing. Conversely, in
studies that used other tasks to test bilinguals that are less proficient in their L2 than their L1,
emotional effects are reduced or absent. For example, bilinguals show reduced skin conductance
responses to L2 childhood reprimands and taboo words in emotion rating tasks (even though
38
explicit ratings show intact emotional word processing; Harris, Ayçiçeği, & Gleason, 2003).
Similarly, bilinguals show reduced activation of emotional meanings in affective priming tasks
(Degner et al., 2012) and implicit affect association tasks (Segalowitz, Trofimovich, Gatbonton,
& Sokolovskaya, 2008).
Another source of complexity regarding L2 emotional word processing concerns different
effects for negative vs. positive valence. Unfortunately, most studies focus on negative
valence—few compare negative, neutral, and positive words to each other, while accounting for
the other linguistic ways that words vary (e.g., frequency, length, etc.). A recent exception is
Conrad, Recio, and Jacobs (2011). They compared words varying on emotional valence using
event-related potentials (ERP) gathered during lexical decisions for two groups of bilinguals
differing in L2 proficiency. Conrad et al. found that more proficient bilinguals showed an
enhanced early posterior negativity and late positive complex in their L1 for negative and
positive words vs. neutral words. These bilinguals showed similar but delayed effects in their
L2. However, less proficient bilinguals showed these ERP modulations in their L2 only for
positive words but not negative words. Thus, bilinguals may treat negative words, but not
positive words, in an unemotional manner in their L2. Moreover, the results in Conrad et al.
(2011) suggest that emotional effects may vary as a function of L2 proficiency. However,
Degner et al. (2012) and Segalowitz et al. (2008), who both found reduced activation of
emotional meanings in L2, found no interactions with continuous measures of L2 proficiency.
Although the findings vary, a common pattern is that bilinguals who are less proficient in
their L2 than their L1 seem to process L2 words in an unemotional manner, and this unemotional
processing may be specific to negative words (Conrad et al., 2011). However, intact emotional
word processing does not appear to be directly predicted by L2 proficiency (Degner et al., 2012;
Segalowitz et al., 2008). Thus, the kind of L2 experience required for intact emotional word
processing remains an open question.
39
One goal in this thesis is to determine whether an embodied approach to word
representation can explain these variable findings. Although embodied theories focus on L1
representation and do not explicitly address L2 semantic representation in bilinguals, it is
possible to derive predictions from previous work. For example, recall that an embodied
approach to word representation assumes that words are grounded in the experiences that cooccur with language use (Vigliocco et al., 2009; Zwaan, 2008). Thus, an embodied perspective
forces us to consider the behavioral ecology (Green, 2011) that bilinguals inhabit, and whether
their ecology affords the emotional embodiment of L2 words. For example, if bilinguals
primarily use their L2 in social contexts that do not provide opportunities for words to co-occur
with emotional experiences, then their L2 words would be emotionally “disembodied”
(Pavlenko, 2012). This idea finds support in sociolinguistic work on language choice. For
example, Dewaele (2004) found that bilinguals curse to express anger using their L1 more than
their L2. Similarly, Pavlenko (2005) found that bilinguals switch to their L1 for emotionally
charged interactions with their romantic partners. Thus, bilinguals may lock their L2 out of
emotional contexts, reducing the probability that L2 words will co-occur with emotional
experiences, leaving them emotionally disembodied.
However, it may also be the case that negative words are more at risk of being
emotionally disembodied than positive words to the extent that bilinguals lock their L2 out of
negative but not positive emotional contexts. Conrad et al. (2011) proposed a similar idea to
explain their findings. They suggested that emotional word processing was intact only for
positive words in their study because a positivity bias ensures that bilinguals use their L2 in
emotionally positive contexts. The idea of a positivity bias is based on work showing that
human communication focuses on emotionally positive exchanges (Boucher & Osgood, 1969).
This idea is also supported by recent work on how emotion regulation varies across social
contexts (Matsumoto, Yoo, Hirayama, & Petrova, 2005). Matsumoto et al. showed that positive
40
emotions are more often up-regulated and less often down-regulated than negative emotions.
Moreover, negative emotions are more often down-regulated with colleagues and strangers vs.
family and friends. Thus, there are fewer contexts that afford opportunities for language to cooccur with emotionally negative experiences than positive experiences, which suggests that it
may be more likely for negative words than positive words to be left emotionally disembodied in
an L2 that gets used only in a subset of the entire space of social contexts. In addition, an
embodied perspective also forces us to consider whether L2 words can be disembodied in terms
of emotional and sensorimotor information, which are reflected by emotional and concreteness
effects, respectively (Kousta et al., 2011). For example, there is no research suggesting that
sensorimotor experiences get locked out of particular social contexts the way emotionally
negative experiences do (Matsumoto et al., 2005). Thus, sensorimotor disembodiment may be
less likely than emotional disembodiment for L2 words, suggesting that concreteness effects
should remain intact in bilinguals.
Eye-Movement Measures of Reading as a Means of Investigating Embodiment
The vast majority of studies on emotional word processing use response-based tasks (e.g.,
lexical decision). Embodied theories are also based on data produced by such tasks. These tasks
have important limitations. For example, they require participants to make overt responses that
are often specific to laboratory experiments (e.g., lexical decisions or color naming in Stroop like
paradigms). These overt responses are not characteristic of natural language processing.
Moreover, words are generally presented without a sentential context in response-based tasks,
making these experiments even more dissimilar from natural language processing. These factors
raise questions about the ecological validity of effects obtained with such tasks. For example, it
is not unlikely that these tasks recruit cognitive processes that do not characterize natural
language processing. Thus, the effects found with such tasks could potentially be artefacts of
response-based paradigms. This possibility undermines theoretical work on embodiment in word
41
representation, assuming that the goal of this work is to produce a complete theory of word
representation, rather than a theory of language processing specific to the laboratory.
Thus, eye-movement sentence reading tasks (Rayner, 1998; 2009) are ideal for the
questions in this thesis. In such tasks, participants read sentences on a computer monitor while
their eye movements are recorded, and their only goal is comprehension (i.e., they are not
required to make any responses while they read sentences). The recording of their eye
movements is completely unobtrusive. Thus, for all intents and purposes, participants read
sentences on the screen as they would any other text.
Research on reading focuses on two components of eye movements: The eye movements
themselves (i.e., saccades) and periods in between saccades when the eyes are relatively
stationary (i.e., fixations; see Rayner, 2009, for a review). Saccades move the eyes from one
word to the next. Most saccades are progressive, which move the eyes to a subsequent word in
the sentence; relatively few are regressive, which move the eyes back to preceding words that the
eyes already moved past (but these increase with text difficulty). Vision is generally suppressed
during saccades (Matin, 1974). Thus, the uptake of visual information (and thus word
processing) for any given word in a sentence generally occurs during fixations, and fixation
durations on a word are used to measure word processing.
Moreover, fixation duration measures can be divided into early measures of word
processing and late measures of word processing (Rayner, 1998, 2009). For example,
researchers typically use first fixation duration (the duration of the first fixation on a word) and
single fixation duration (the fixation duration in cases where the word was fixated exactly once)
to measure the earliest stages of word processing during direct fixations on words, given that
they reflect processing time during the first (or only) fixation on a word. Researchers also
measure early stages of word processing using gaze duration (the sum of the durations of all
fixations made during the first pass, before the eyes left the word region), though gaze is a
42
relatively later measure of early word processing compared to first fixation duration and single
fixation duration, given that gaze duration includes first-pass refixations on target words.
Another measure of first-pass processing that is later in the time course of reading is go-past time
(the sum of the durations of all fixations on the word from the point when the word is first
fixated up until the eyes move past the word to the right). Finally, researchers can also measure
late stages of word processing using second pass time (the sum of the durations of all fixations
on the word made after the first pass) and total reading time (the sum of all fixation durations on
a word, including fixations following regressive saccades back to the word). Thus, eyemovement sentence reading tasks can be used to determine the time course of experimental
effects more precisely than in response-based tasks.
Another important difference between response-based tasks and eye-movement sentence
reading tasks is that individuals have multiple opportunities to process words in sentence reading
tasks, which is part of the natural experience of reading sentences. For example, words can be
processed when they are directly fixated, which brings them into foveal vision (i.e., 2 degrees
around fixation), where visual acuity is highest. However, upcoming words that have not yet
been fixated (in the parafovea, which extends from the fovea up to approximately 5 degrees on
either side of fixation) can also be processed while preceding words are fixated. This has been
demonstrated in research on word skipping. People skip content words and function words about
15% and 65% of the time while reading, respectively, without disrupting comprehension
(Rayner, 2009). Thus, people can identify words when they are to the right of fixation in the
parafovea, or when they are brought into foveal vision with direct fixations. However, many
words are processed both foveally and parafoveally, which has been demonstrated in research on
preview benefits. Preview benefits refer to shorter fixation durations on words that are
parafoveally processed before being fixated, which can be measured using the gaze-contingent
boundary paradigm (Rayner, 1975). Previous work shows that fixation durations on target words
43
are approximately 30-50 ms shorter when a valid parafoveal preview is available compared to
when that preview is denied (e.g., by presenting a nonword instead of the actual word before it is
fixated; for a review, see Rayner, 2009).
Some work based on the boundary paradigm is the subject of controversy. For example,
some researchers find that the frequency of words is parafoveally processed during reading
before those words are directly fixated (Kennnedy & Pynte, 2005; Kliegl, Nuthmann, & Engbert,
2006; Risse & Kliegl, 2012; Wotschack & Kliegl, 2013). However, the mechanisms underlying
such effects are the subject of intense debate (see Drieghe, 2011; Schotter, Angele, & Rayner,
2012). Whatever the mechanisms underlying such effects turn out to be, these studies add to the
body of evidence that word processing during reading is not strictly localized in space. Thus,
eye-movement measures provide insight into both the time course of effects, thanks to their
exquisite temporal resolution, and the point in space when word processing begins, thanks to
their exquisite spatial resolution. This thesis capitalizes on both of these strengths.
The Present Dissertation
A fundamental question in the study of language concerns whether the semantic meanings
of words are grounded in representations that directly encode our experiences, as proposed in
embodied approaches to cognition (e.g., Barsalou, 1999; Glenberg, 1997; Lakoff & Johnson,
1980), or represented by some kind of “mentalese” or language of thought that has been stripped
of such experiential knowledge, as in modular theories of cognition (e.g., Fodor, 1975, 1983).
Although the traditional approach assumes modularity, embodied approaches have increasingly
gained mindshare over the last 15 years. However, an often-cited problem with embodied
approaches is how to ground the meanings of abstract words which have referents that are not
directly experienced by the senses. A recent proposal is that emotional experiences play a pivotal
role in word representation, and are particularly important for grounding the meanings of abstract
words. However, there are several open questions about the role of emotion. First, how does
44
emotional embodiment, as indexed by emotional valence, influence word processing for words
that vary in terms of their sensorimotor embodiment, as indexed by concreteness, and their
linguistic contributions to word representation, as indexed by word frequency? Second, does
emotional embodiment also play a fundamental role in the representation of L2 word
representations? Third, does embodiment based on negative vs. positive emotion influence word
processing differently? These questions are addressed in this thesis with eye-movement
measures of sentence reading.
This thesis makes an original and novel contribution to the literature in several ways.
First, naturalistic eye-movement sentence reading tasks were used in each study. Thus, this
thesis addresses whether emotional effects on word processing and embodied theories are
ecologically valid, which is an important contribution given that previous work on these topics is
mostly based on laboratory type response-based tasks. Second, linear mixed models (LMMs;
Bates et al., 2008) were used to analyze the data in each study, which provides several
advantages over traditional statistical methods. For example, LMMs account for multiple
sources of variance in the data/correlations within clusters of data (e.g., observations for the
same participants and/or the same experimental items). This property increases the
generalizability of statistical tests in LMMs. Moreover, this property makes it unnecessary to
average observations over participants/items, which is the general approach in the analysis of
variance (ANOVA). Third, this property also makes it possible to examine how experimental
effects vary as a function of continuous individual differences across participants (which is done
in Chapters 2 and 3). Thus, this thesis also addresses timely questions about individual
differences, which are increasingly receiving attention in psychology and beyond.
This thesis consists of five chapters. Chapter 1, which includes this section, consists of
an introduction to embodied approaches to word representation, a review of research on L1
emotional word processing and L2 emotional word processing, and a brief introduction to eye45
movement measures of reading. Chapters 2, 3, and 4 consist of original research, following a
manuscript-based format, and the goal of each chapter is described below. Chapter 5 consists of
a summary of the findings, and a discussion of the findings, including implications for the work
reviewed above as well as future directions.
Chapter 2 (Sheikh & Titone, 2013, Emotion): This chapter presents an eye-movement
sentence reading study that examined how native speakers of English process English target
words embedded in emotionally neutral sentences. It was predicted that negative and positive
valence would both generally facilitate target word processing when concreteness and word
frequency were low, given the reasoning that embodied information should functionally
modulate word processing primarily when sensorimotor and linguistic aspects of representation
are insufficient for generating a response. The same effects were generally expected for
embodiment based on negative and positive emotion, given the reasoning that the behavioral
approach and avoidance systems activated by them, respectively, are equally important to
survival. However, it was also predicted that the effects of negative and positive valence may
differ as a function of individual differences in sensitivity to emotional and sensorimotor
information. The novel contribution of this study is that it demonstrates how emotional,
sensorimotor, and linguistic aspects of representation jointly influence word recognition.
Moreover, this study investigates how individual differences in sensitivity to emotional and
sensorimotor information influence eye movements during reading, which are also novel
contributions to research on embodiment.
Chapter 3 (Sheikh & Titone, 2015a, Cognition & Emotion): This chapter presents an
eye-movement sentence reading study that examined how English target words embedded in
emotionally neutral sentences are processed by bilinguals reading in their L2 (L2 readers), who
are French-English bilinguals (native speakers of French). The stimuli are the same as in the
study presented in Chapter 2, which permitted a direct comparison with the native speakers of
46
English in that study reading in their L1 (L1 readers). As in Chapter 2, it was predicted that
emotion would facilitate target word processing when concreteness was low. However, it was
predicted that emotional facilitation may not be limited to low-frequency words in L2 readers, as
in L1 readers, given the reasoning that bilinguals experience words less frequently overall in
their non-dominant, less proficient language. Moreover, it was predicted that word processing
facilitation by positive valence would manifest across early and late measures of reading, and
that facilitation by negative valence may be more limited, given the reasoning that it may be
more difficult to activate embodied information for L2 words that are disembodied in terms of
emotionally negative experiences. Finally, emotional and concreteness effects were expected to
be differentially modulated by individual differences in L2 proficiency. The novel contribution
of this study is that it investigates how bilinguals process emotional words in sentences in their
L2 using eye-movement measures of sentence reading. It also investigates how emotional,
sensorimotor, and linguistic aspects of representation jointly influence L2 word recognition in
bilinguals, which are also novel contributions to research on embodiment.
Chapter 4 (Sheikh & Titone, 2015b, submitted to Emotion): This chapter presents an
eye-movement sentence reading study that used the gaze-contingent boundary paradigm to
examine how native speakers of English process English target words embedded in emotionally
neutral sentences. The study in this chapter follows up on a salient difference between negative
and positive valence across the studies in Chapters 2 and 3. Specifically, emotional facilitation
by positive valence was found for early eye-movement measures in both studies. However,
emotional facilitation by negative valence for early eye-movement measures was found only for
the L1 readers in Chapter 2. A careful analysis of other work in the literature suggests that the
early effects of negative vs. positive valence may differ when preceding regions of text are more
difficult to process, which also suggests that differences between negative vs. positive valence
may depend on whether readers parafoveally process emotional words, before they are directly
47
fixated. Thus, a new set of sentences was created for the study in Chapter 3, and it was predicted
that early emotional facilitation would be limited to positive valence, and that this facilitation
would vary as a function of word frequency and whether readers were able to parafoveally
process target words. The novel contribution of this study is that it investigates how emotional
words are processed in English sentences using eye-movement measures of reading and the
boundary paradigm.
48
CHAPTER 2:
Sensorimotor and Linguistic Information Attenuate Emotional Word Processing Benefits:
An Eye-Movement Study
(Sheikh & Titone, 2013, Emotion)
49
Abstract
Recent studies report that emotional words are processed faster compared to neutral
words, though emotional benefits may not depend solely on words’ emotionality. Drawing on an
embodied approach to representation, we examined interactions between emotional,
sensorimotor, and linguistic sources of information for target words embedded in sentential
contexts. Using eye movement measures for 43 native English speakers, we observed emotional
benefits for negative and positive words and sensorimotor benefits for words high in
concreteness, but only when target words were low in frequency. Moreover, emotional words
were maximally faster than neutral words when words were low in concreteness (i.e., highly
abstract), and sensorimotor benefits occurred only when words were not emotionally charged
(i.e., emotionally neutral). Furthermore, emotional and concreteness benefits were attenuated by
individual differences that attenuate and amplify emotional and sensorimotor information,
respectively. Our results suggest that behavior is functionally modulated by embodied
information (i.e., emotional and sensorimotor) when linguistic contributions to representation are
not enhanced by high frequency. Furthermore, emotional benefits are maximal when words are
not already embodied by sensorimotor contributions to representation (and vice versa). Our work
is consistent with recent studies that suggest that abstract words are grounded in emotional
experiences, analogous to how concrete words are grounded in sensorimotor experiences.
50
Introduction
People are exquisitely sensitive to emotionally salient stimuli, such as angry or happy
faces (see Dolan, 2002, for a review). Interestingly, emotional words, such as sex, can also
undergo enhanced processing (e.g., Anderson & Phelps, 2001), although, unlike faces, they are
symbolic stimuli that must be learned through experience. However, several questions about
emotional word processing remain unclear, which include whether emotional words are also
prioritized during natural language processing, whether emotional word processing advantages
are modulated by other key lexical variables (e.g., frequency and concreteness), and whether
individual differences in emotional or sensorimotor processing alter the basic pattern of emotion
word processing effects. In this study, we use eye movement recordings to investigate whether
the earliest stages of natural language processing are faster for emotionally charged words during
natural sentence reading. We were particularly interested in clarifying what key linguistic,
semantic, and individual difference variables modulate emotional processing benefits.
Emotional processing benefits are often thought to reflect neural mechanisms that
enhance sensory and behavioral responses for emotionally negative and positive stimuli, thus
enabling rapid responses to aversive events (e.g., avoiding violence) and appetitive events (e.g.,
approaching mates) (Lang & Bradley, 2010; Lang, Bradley, & Cuthbert, 1990, 1997). Although
several early studies showed that processing was slower for negative words compared to neutral
or positive words using lexical tasks (e.g., Algom, Chajut, & Lev, 2004; MacLeod, Tata, &
Mathews, 1997; Wentura, Rothermund, & Bak, 2000), many of these studies did not control key
lexical attributes known to influence word recognition, such as word frequency and length
(Larsen, Mercer, Balota, & 2006; see also Kousta, Vinson, & Vigliocco, 2009, for an expanded
list of lexical variables). Recent studies using more tightly controlled stimulus materials show
that processing is generally faster for both negative and positive words compared to neutral ones
(Holt et al., 2006; Kousta et al., 2009; Kuchinke, Võ, Hofmann, & Jacobs, 2007; Long & Titone,
51
2007, Méndez-Bértolo, Pozo, & Hinojosa, 2011; Scott, O'Donnell, Leuthold, & Sereno, 2009;
Scott, O'Donnell, & Sereno, 2012), though some studies have found differences between positive
and negative words as a function of frequency (Kuchinke et al., 2007; Scott et al., 2009; Scott et
al., 2012).
Recent work appealing to embodied theories of cognition (e.g., Barsalou, 1999; Barsalou,
2008; Barsalou, Simmons, Barbey, & Wilson, 2003; Decety & Grezes, 2006; Gibbs, 2006;
Rizzolatti & Craighero, 2004) has the potential to suggest a more nuanced view regarding the
role of emotionality on word processing (e.g., Vigliocco, Meteyard, Andrews, & Kousta, 2009).
Embodied theories hypothesize that knowledge, including word representations, is grounded in
experiential information (Barsalou, 1999). For example, with regard to words, Vigliocco et al.
(2009) posited that one experiential source of information consists of sensorimotor
representations of sensory events (e.g., vision) and actions (e.g., movements) in the external
world, and a second experiential source of information consists of emotional representations of
states in the inner world (e.g., negative and positive affect). In addition to these embodied
sources of information, Vigliocco et al. (2009) also posit that knowledge is represented by purely
linguistic sources of information (e.g., verbal associations and syntactic information) (see also
Barsalou, Santos, Simmons, & Wilson, 2008; Louwerse & Jeuniaux, 2008; Louwerse &
Jeuniaux, 2010; Zwaan, 2008).
This work suggests that these three foundational sources of information, sensorimotor,
emotional, and linguistic, have the potential to simultaneously modulate the conditions under
which processing benefits emerge for emotional words. For example, previous work suggests
that embodied information does not affect processing when linguistic information is sufficient
for a response (Barsalou et al., 2008). This prediction is supported by work showing that
emotional processing benefits are reduced for high vs. low frequency words, given the plausible
assumption that word frequency makes a linguistic contribution to representation (Kuchinke et
52
al., 2007; Mendez-Bertolo et al., 2011; Scott et al., 2009; Scott et al., 2012).
As well, a second potential implication of embodied views, setting aside the idea of
linguistic contributions to word processing, is that emotional benefits can also be modulated by
sensorimotor contributions to word processing. For example, recent work shows that words that
are high in concreteness are associated more with sensorimotor information than words that are
low in concreteness, that is, abstract words (Kousta, Vigliocco, Vinson, Andrews, & Del Campo,
2011). While the standard concreteness effect in language studies is that concrete words are
processed more rapidly than abstract words (Binder, Westbury, McKiernan, Possing, & Medler,
2005; Bleasdale, 1987; de Groot, 1989; James, 1975; Kroll & Merves, 1986; Schwanenflugel,
Harnishfeger, & Stowe, 1988; Schwanenflugel & Stowe, 1989; Whaley, 1978), a recent large
scale analysis of thousands of words showed, counterintuitively, that abstract words were
processed faster than concrete words, presumably because concrete words tend to be less
emotional and abstract words tend to be more emotional in general (Kousta et al., 2011). A
general relationship between concreteness and emotionality is also consistent with work showing
that the least concrete words (i.e., words that are highly abstract) more likely refer directly to
emotional states (e.g., anger) (Altarriba, Bauer, & Benvenuto, 1999), and have more emotional
features than concrete words (Barsalou & Wiemer-Hastings, 2005).
Thus, one open question concerns how emotionality affects the processing of words that
vary in their sensorimotor contribution (i.e., concreteness) when emotionality is not confounded
with concreteness. While these studies suggest that concreteness is correlated with emotionality
in large sets of words, it is possible to orthogonally manipulate concreteness and emotionality in
subsets of words. Thus, some concrete words are emotionally charged (e.g., snake) and some
abstract words are emotionally neutral (e.g., dozen). To our knowledge, no previous study has
examined the joint effects of frequency, emotional valence, and concreteness. We believe an
embodied view would predict that the contribution of emotionality to word processing would be
53
more likely for words that do not already enjoy processing advantages based on sensorimotor
information, that is, abstract words, or on linguistic information, that is, low frequency words.
Thus, an embodied view predicts maximal emotional processing benefits for low frequency
abstract words.
A second open question that emerges from the embodied view concerns individual
differences in how people may differentially weight emotional or sensorimotor information. For
example, alexithymia is a personality trait that is often described as involving a difficulty
identifying and describing feelings combined with a hypersensitivity to sensorimotor information
from the external world (Bagby, Parker, & Taylor 1994; Bagby, Taylor, & Parker, 1994; Sifneos,
1973). Consistent with this description, people who score high on alexithymia show reduced
interference from negative words in color naming Stroop paradigms (Mueller, Alpers, & Reim,
2006), recall fewer emotional words on memory tasks (Luminet, Vermeulen, Demaret, Taylor, &
Bagby, 2006), and exhibit reduced priming by emotional words (Goerlich, Witteman, Aleman, &
Martens, 2011). Alexithymia may also differentially influence processing for emotionally
positive stimuli compared to negative stimuli. For example, the effect of alexithymia on memory
for negative stimuli is conditional on individual current positive and negative state affect,
optimism, and depression; however, the effect for positive stimuli does not seem to depend on
these additional variables (Luminet et al., 2006; see also Goerlich et al., 2011). People high on
alexithymia also show hypersensitivity to sensorimotor information in studies on touch (Sivik,
1993), pain (Nyklicek & Vingerhoets, 2000), and acoustic stimuli (Schafer et al., 2007).
However, previous studies have not investigated differences with respect to the emotional vs.
sensorimotor (i.e., concrete vs. abstract) qualities of symbolic stimuli such as words. Here, we
believe that the embodied view predicts that high alexithymia may simultaneously attenuate the
influence of emotional information and amplify the influence of sensorimotor information.
54
The Present Study
Taken together, the existing literature suggests that processing benefits for negative and
positive words compared to neutral words may be determined in part by frequency and
concreteness, and also by individual differences in how people weight emotional vs.
sensorimotor information, as in alexithymia. Here, we investigated these ideas using eye
movement measures of natural reading (Rayner 1998; 2009), specifically focusing on three key
questions: (1) Are emotional words prioritized when people read naturally for comprehension?
(2) Are emotional word processing advantages modulated by linguistic and sensorimotor
attributes of words (frequency and concreteness)? and (3) Do individual differences in emotional
or sensorimotor processing reflected in alexithymia alter the basic pattern of emotional
processing benefits? An embodied perspective suggests that emotional processing benefits
should be most likely to emerge when particular words do not already enjoy a processing
advantage from high frequency or high concreteness. As well, an embodied perspective suggests
that emotional and concreteness benefits should be reduced by individual differences in
alexithymia, which attenuates the influence of emotional information and amplifies the influence
of sensorimotor information.
Thus, we embedded target words, comprised exclusively of nouns that varied on
emotional valence, frequency, and concreteness, in semantically neutral sentences. Sentences
were presented in their entirety while participants’ eye movements were recorded as they read
naturally with no goal other than comprehension. We analyzed both early-stage (i.e., gaze
duration) and late-stage (i.e., second pass time) reading measures on the target words (Rayner,
1998; 2009). We selected gaze duration because it is more appropriate than alternative measures
(first fixation duration and single fixation duration) given our experimental manipulations.
Specifically, single fixation duration would preferentially exclude low frequency words that are
likely to be fixated more than once. Similarly, first fixation duration would underestimate
55
reading times for low frequency words for the same reason. We selected second pass time
because it is a relatively pure measure of late-stage processing compared to alternative measures
(e.g., total reading time).
We predicted that reading time benefits for emotional words would be found for earlystage processing to the extent that emotionality affects the earliest stages of comprehension and
for late-stage processing. Moreover, we predicted faster reading times for low frequency negative
and positive words compared to neutral words (Kuchinke et al., 2007; Scott et al., 2009; Scott et
al., 2012), based on the idea that embodied emotional information is less likely to influence
processing when linguistic information is readily available due to high frequency (Barsalou et
al., 2008; see also Juhasz, Yap, Dicke, Taylor, & Gullick, 2011). We further hypothesized that
processing benefits for negative and positive words compared to neutral words would be
maximal when words were low in concreteness (i.e., low in sensorimotor associations) (Kousta
et al., 2011). For words high in concreteness, we predicted that the benefit for emotional
compared to neutral words would be reduced or eliminated because these neutral words should
be processed faster because their representations are enriched by sensorimotor information
(Binder et al., 2005; Bleasdale, 1987; de Groot, 1989; James, 1975; Kroll & Merves, 1986;
Schwanenflugel et al., 1988; Schwanenflugel & Stowe, 1989; Whaley, 1978). Conversely,
processing advantages for concrete words rich in sensorimotor information should be maximal
when words are emotionally neutral and low in frequency.
With respect to alexithymia, we predicted that reading time benefits for emotional words would
be reduced or eliminated when alexithymia was high (Goerlich et al., 2011; Luminet et al., 2006;
Mueller et al., 2006). We also predicted that alexithymia would reduce any differences between
words that are low vs. high in concreteness given that people who score high on alexithymia
would likely treat all words as functionally more concrete (Nyklicek & Vingerhoets, 2000;
Schafer et al., 2007; Sivik, 1993).
56
Method
Participants
Forty-three native English speakers (mean age = 24.14, SD = 5.14, range = 19-35) were
tested from McGill University and the greater Montreal community. They reported no
neurological disorders, normal or corrected-to-normal vision, and were paid $10.00 per hour or
received course credit for compensation.
Materials and Design
We used negative, neutral, and positive target words (156 total) that were grouped into 52
triplets. The words were selected from a set of over 1200 words, which were normed for
emotional valence and arousal by Kousta, et al. (2009). The stimulus specifications are presented
in Table 1. Negative, neutral, and positive words had mean valence ratings of 3.11 (SD = .68),
5.33 (SD = .52), and 7.09 (SD = .44), respectively. Targets were low (M = 7.52 [SD = 1.12]) or
high (M = 10.12 [SD = .88]) on log frequency. The emotional categories were balanced—within
the low and high frequency levels—on frequency, concreteness, and length; though length
differed across low and high frequency items (length was shorter for high frequency compared
low frequency words). However, we ensured that these length differences did not influence our
hypothesis tests by including length as a covariate in all analyses. The emotional categories
differed on valence (negative < neutral < positive). For arousal, negative and positive words did
not differ, but both were higher on arousal compared to neutral words, which is typical in studies
of emotional words. The target words were embedded in sentences which forced a noun
interpretation for all target words by having targets always follow and precede the words ‘the’
and ‘that’, respectively (e.g., Jim knew that the dishonesty/somersault/happiness that he
demonstrated would affect the judges.).
57
Table 1. Mean (Standard Deviations) For Lexical Attributes For Target Words And Their Latent Semantic Analysis
(LSA) Values Within Sentence Frames
Low frequency
High frequency
Negative
Neutral
Positive
Negative
Neutral
Positive
Log frequencya1 7.58 (1)
7.18 (1.27)
7.86 (1.03)
9.89 (0.77)
10.31 (0.91)
10.11 (0.91)
b
Valence
3.15 (0.58)
5.28 (0.51)
7.11 (0.42)
3.05 (0.85)
5.39 (0.54)
7.08 (0.45)
Arousalb
5.13 (0.8)
4.36 (0.87)
5.42 (1.07)
5.25 (0.65)
4.54 (0.73)
5.54 (0.88)
c
Concreteness
429.62 (106.9) 463.6 (138.2) 459.14 (125.8) 445.4 (120.85) 471.93 (116.24) 468.42 (129.05)
Length
6.31 (2.32)
7.2 (3.52)
7 (2.14)
5.9 (2.34)
4.81 (2.37)
5.39 (1.75)
LSA value
-0.005 (0.06)
0.013 (0.06) 0.005 (0.06)
0.004 (0.05)
0.009 (0.04)
0.01 (0.06)
a
The English Lexicon Project (Balota et al., 2007)
b
Kousta et al. (2009)
c
MRC Psycholinguistic Database (Coltheart, 1981)
1
An ANOVA showed that there were no significant differences on raw HAL frequency between negative, neutral, and positive words for low
frequency words (p > .25) or high frequency words (p > .46).
58
An important methodological problem that arises when comparing processing for target
words embedded in sentences, which does not typically arise when comparing isolated words in
single-word paradigms, is the need to control for potential differences in “fit” between target
word-sentence frame pairings. For example, a particular target word may be more or less
predictable or plausible in a given sentence relative to another target word that it is compared
with (e.g., friends vs. shirts in “While I was visiting my home town, I had lunch with several old
friends/shirts”). The issue of fit is particularly problematic when comparing emotional vs. neutral
words embedded in neutral sentences because negative and positive words, by virtue of their
emotionality, are emotionally incongruent with emotionally neutral sentential frames. For
example, Holt, Lynn, & Kuperberg (2009) showed in an event-related potential (ERP) study that
the N400 component was larger for emotional compared to neutral words when embedded in the
same emotionally neutral sentences even after analyzing only those items that were matched on
predictability and plausibility according to human ratings.
We took the following steps to ensure that our findings would not be influenced by
differential fit between sentences and target words. First, we created sentential contexts in which
all target words were completely unpredictable and without semantic associations to the sentence
frames, rather than equally predictable and semantically associated (see Appendix A). We used
latent semantic analysis (LSA) to confirm that there were no differential associations between
negative, neutral, and positive words and the sentential contexts (see Table 1). The LSA values
did not differ as function of valence (p > .11), frequency (p > .51), or their interaction (p > .57).
Moreover, we used an experimental design that, when combined with linear mixed model
analyses (LMM; Bates, Maechler, & Dai, 2009), allows us to statistically control for differences
in fit by modeling the random effects for the target word-sentence pairings. Specifically, we
grouped negative, neutral, and positive words into triplets and wrote three sentence frames for
each triplet. The sentence frames formed well-formed sentences when combined with their
59
triplets’ negative, neutral, or positive word. In a given experimental list, each target was
presented once in a different sentence frame. However, target words were combined with
different sentence frames in the remaining two lists (see Table 2 for a concrete example). Thus,
negative, neutral, and positive words in a given triplet appeared in the same sentence frames
across counterbalance lists (i.e., target word and sentence frame were fully crossed across
participants). We then used the properties of this fully crossed design to model the variability for
the set of sentences in which a triplet’s target words appeared by adding a random effect for
triplets (in addition to random effects for target words and participants). Thus, we statistically
control for any potential differences in fit across negative, neutral, and positive words whether
driven by plausibility, predictability, or some unknown factor(s).
Table 2. Sentence Frames Were Combined With A Single Target Word (Between Asterisks) In A
Given List But Were Rotated Around Their Triplets’ Negative, Neutral, And Positive Words So That
Target Words Were Fully Crossed With Sentence Frames Across Participants. Note That Target
Words Were Not Presented With Asterisks During The Actual Experiment.
Counterbalance
Sentence
List 1
The art teacher presented the *smoke* that the students were going to paint.
The news report mentioned the *hill* that was discussed at work.
The school paper reviewed the *drink* that had caused concern among
students.
List 2
The art teacher presented the *drink* that the students were going to paint.
The news report mentioned the *smoke* that was discussed at work.
The school paper reviewed the *hill* that had caused concern among students.
List3
The art teacher presented the *hill* that the students were going to paint.
The news report mentioned the *drink* that was discussed at work.
The school paper reviewed the *smoke* that had caused concern among
students.
Alexithymia was assessed using the 20-item Toronto Alexithymia Scale (TAS-20; Bagby,
Parker, & Taylor 1994), which is the most widely used measure of this personality trait. The total
alexthymia score on the TAS-20 reflects variability on three constructs: difficulty identifying
feelings, difficulty describing feelings, and externally oriented thinking. Each item is rated on a
60
5-point scale ranging from “strongly agree” (5) to “strongly disagree” (1). We used the total
alexithymia score in our analyses.
A modified Language Experience and Proficiency Questionnaire (LEAP-Q; Marian,
Blumenfeld, & Kaushanskaya, 2007) was used to verify that all participants were native English
speakers.
Apparatus and Procedure
Eye movements were recorded using an Eyelink 1000 eye tracker that sampled eye
position every millisecond. Viewing was binocular but only the right eye was recorded.
Sentences were displayed on a 21” Viewsonic CRT monitor with a refresh rate of 144 Hz, a
resolution of 1024 × 768 in yellow 10-point Monaco font on a black background, with 3
characters subtending approximately 1 degree of visual angle. Eye movements were calibrated
using a 3-point grid. Target words and sentence frames were presented once in each of three
counterbalance lists. Across lists, each sentence frame was paired once with its triplet's negative,
neutral, and positive word. Twenty-one percent of trials were followed by a yes or no question
about the immediately preceding sentence to ensure participants read sentences for
comprehension. The LEAP-Q and TAS-20 were administered after subjects completed the
reading task.
Results
Our analysis focused on standard reading time measures that reflect early- and late-stage
processing for the target word region (Rayner 1998; 2009). We used gaze duration (the sum of
the duration of all fixations made during the first pass, before the eyes left the region) to measure
early processing stages, and second pass time (the sum of the durations of all fixations made
after the first pass) to measure late processing stages. We supplement our analyses of fixation
durations with analyses of fixation probability and regression probability for the target region,
though we did not form hypotheses for these measures. Means and standard deviations broken
61
down by reading measure, valence, frequency, and concreteness are presented in Table 3
(concreteness, which was analyzed continuously, was median split in each valence-frequency
cell in order to produce this table). Mean accuracy on the comprehension questions was 87.4%
(SD = 5.22) and ranged from 78.6-100%.
Table 3. Mean (Standard Deviations) For Eye Movement Measures For Target Words
Low frequency
Low concreteness
High concreteness
Negative Neutral
Positive
Negative Neutral
Positive
Gaze duration
277 (112) 330 (161) 286 (126) 273 (120) 275 (121) 266 (100)
Second pass time
307 (168) 383 (258) 321 (186) 288 (157) 334 (215) 275 (186)
Fixation probability
0.88
0.88
0.89
0.81
0.79
0.86
Regression probability 0.19
0.16
0.17
0.16
0.2
0.15
High frequency
Low concreteness
High concreteness
Negative Neutral
Positive
Negative Neutral
Positive
Gaze duration
273 (107) 253 (96) 252 (104) 254 (91) 255 (98) 241 (84)
Second pass time
269 (143) 334 (231) 275 (150) 263 (126) 300 (174) 274 (156)
Fixation probability
0.77
0.77
0.8
0.8
0.68
0.76
Regression probability 0.18
0.18
0.13
0.15
0.17
0.18
For the analysis of fixation duration measures, 25% of the trials were discarded due to
track losses, absence of fixations on targets, or when observations were shorter than 80 msec.
From the remaining observations, visually identified outliers were eliminated (gaze durations
above 1000 msec [0.3%] and second pass times above 1500 msec [0.72%]). The remaining
observations were log-transformed and analyzed in R (R Development Core Team, 2010) with
linear mixed models (LMM) using the lme4 package (Bates, Maechler, & Dai, 2009), which
permit inclusion of subjects and items as crossed random effects that provide advantages over
traditional analyses (Baayen, Davidson, & Bates, 2008). We applied an inverse transformation
when generating plots to present partials effects in non-log space. We also used LMMs so that
we could investigate continuous individual differences over people and items in the same model
for our alexithymia analyses. To test the joint effects of valence, frequency, and concreteness, the
62
eye movement data were fit to models that included predictors for valence (negative vs. neutral
vs. positive), frequency (low vs. high), and concreteness (continuous), and the interactions of
these variables as fixed effects and subject, item, and triplet as random effects. Note that we
present an analysis using a categorical frequency variable for the sake of easy exposition.
However, we analysed frequency both categorically and continuously and the pattern of effects is
the same either way. To test the effect of alexithymia, we added a main effect for the continuous
total alexithymia score on the TAS-20 and its interactions with the preceding fixed effect
predictors.
We also include a covariate for length (continuous) in all models to insure that the pattern
of results does not reflect differences in length between low and high frequency words. For the
sake of brevity, we present the model outputs for the length covariate together with the output for
the other predictors in the tables below, but we do not describe the covariate in the Results
section as it is not theoretically important for our hypotheses.
All continuous predictors were standardized. We report for the fixed effects the
coefficient and standard error estimates as well as p values estimated from Markov chain Monte
Carlo (MCMC) simulations, which are preferred for LMM analyses (see Baayen, 2008).
Generalized linear mixed models (GLMM) with predictors identical to those described above
were fit to the binary dependent variables fixation probability and regression probability. For the
GLMM analyses, we report for the fixed effects the coefficient and standard error estimates as
well as p values for the normal distribution.
Gaze Duration
We modeled gaze durations to test our hypothesis that early-stage emotional word
processing advantages are modulated by linguistic and sensorimotor attributes of words (i.e.,
frequency and concreteness). As can be seen in the LMM output (Table 4), the main effects show
that early processing was faster for positive words (b = -0.07, SE = 0.02, p < .01) compared to
63
neutral words, and for high frequency words compared to low frequency words (b = -0.08, SE =
0.03, p < .01), and for words higher in concreteness (b = -0.03, SE = 0.02, p < .05). There was no
difference between negative and positive words, which we tested by setting negative valence as
the model baseline (all ps > .21). Significant interactions emerged between valence and
concreteness for negative words (b = 0.06, SE = 0.02, p < .01) and for positive words (b = 0.06,
SE = 0.02, p < .01); between frequency and concreteness (b = 0.06, SE = 0.02, p < .05); and
between valence, frequency, and concreteness for negative words (b = -0.1, SE = 0.04, p < .01)
and positive words (b = -0.08, SE = 0.03, p < .05). The partial effects for the valence, frequency,
and concreteness interaction are plotted in Figure 1.
64
Table 4. Interaction between Valence (Negative [Neg], Neutral [baseline], Positive [Pos]), Frequency (Freq: high, low [baseline]), and
Concreteness (Conc: continuous) With Covariates for Length (continuous)
Gaze duration
Second pass time
Fixation probability
Regression probability
b
SE
p
b
SE
p
b
SE
p
b
SE
p
(Intercept)
5.59 0.03 0.0001*** 5.65 0.04 0.0001*** 1.78 0.13 0.0000*** -1.86 0.19 0.0000***
Neg
-0.04 0.02 0.0728
-0.12 0.05 0.0226*
0.13 0.13 0.3382
-0.05 0.15 0.7706
Pos
-0.07 0.02 0.0026**
-0.12 0.05 0.0184*
0.09 0.16 0.5772
-0.12 0.17 0.462
Freq
-0.08 0.03 0.0026**
-0.03 0.06 0.637
-0.07 0.13 0.5922
0.12 0.19 0.5276
Conc
-0.03 0.02 0.0444*
-0.04 0.03 0.2376
0.09 0.09 0.3485
0.08 0.12 0.49
Length
0.07 0.01 0.0001*** 0.07 0.02 0.0006*** 1
0.06 0.0000*** 0.11 0.08 0.1479
Neg by Freq
0.05 0.04 0.1724
-0.05 0.08 0.5194
-0.19 0.18 0.2953
-0.06 0.24 0.8147
Pos by Freq
0.03 0.03 0.314
0
0.08 0.975
-0.1
0.19 0.6059
-0.1
0.24 0.662
Neg by Conc
0.06 0.02 0.0066**
0.04 0.05 0.3902
-0.09 0.13 0.4945
-0.12 0.16 0.4321
Pos by Conc
0.06 0.02 0.0076**
0.02 0.05 0.7584
0.07 0.15 0.6653
-0.27 0.15 0.0832
Freq by Conc
0.06 0.02 0.0168*
0.05 0.05 0.381
-0.1
0.12 0.4254
0
0.18 0.9892
Neg by Freq by Conc
-0.1
0.04 0.0040**
-0.06 0.08 0.497
0.28 0.18 0.1281
-0.08 0.24 0.7392
Pos by Freq by Conc
-0.08 0.03 0.0114*
0
0.07 0.991
-0.02 0.19 0.9076
0.48 0.22 0.0308*
* p < 0.05, ** p < 0.01, *** p < 0.001
65
Figure 1. Partial effects for interaction between valence, frequency, and concreteness showing
faster processing for negative and positive words compared to neutral ones when emotional
words are low in frequency and concreteness
To test our hypothesis that embodied emotional and sensorimotor information do not
produce behavioral effects when words have rich associations with linguistic information (i.e.,
high frequency words), we fit separate models to the gaze duration data for low and high
frequency items. These analyses show that the three-way interaction between valence, frequency,
and concreteness was driven by faster processing for negative words and positive words
compared to neutral words when frequency and concreteness were low, and faster processing for
words higher in concreteness when frequency was low and valence was neutral. However, there
were no effects when frequency was high (all ps > .09). Thus, the three-way interaction between
valence, frequency, and concreteness for gaze durations reflected faster early processing for
negative and positive words compared to neutral ones when frequency and concreteness were
66
low, and faster processing for concrete compared to abstract words when frequency was low and
words were emotionally neutral. Note that we analyzed a second measure of early-stage
processing, first fixation duration, and the pattern of effects was the same as gaze duration. For
the sake of brevity, we do not discuss first fixation duration any further.
Second Pass Time
We modeled second pass times to test the hypothesis that late-stage emotional word
processing advantages were modulated by linguistic and sensorimotor attributes of words (i.e.,
frequency and concreteness). As can be seen in Table 4, there were significant main effects for
valence that showed that late-stage processing was faster for negative words (b = -0.12, SE =
0.05, p < .05) and positive words (b = -0.12, SE = 0.05, p < .05) compared to neutral words. A
model with negative valence as the baseline indicated that there was no difference between
negative and positive words (all ps > .62). No other effects were significant. Thus, the three-way
interaction between valence, frequency, and concreteness observed during early-stage processing
stages measured by gaze durations did not occur at later processing stages measured second pass
time.
Fixation Probability and Regression Probability
We supplement our main analyses with analyses of fixation probability and regression
probability, though we did not formulate any hypotheses with respect to these measures. As can
be seen in Table 4, there were no significant effects for fixation probability. For regression
probability, there was a single significant interaction effect between valence, frequency, and
concreteness for positive words (b = 0.48, SE = 0.22, p < .05). The coefficients in this model
suggest that the probability of making regressions was reduced for positive words high in
concreteness, but that this effect was attenuated for high frequency positive words. However,
when we fit the low and high frequency data to separate models to help explain the three-way
interaction, there were no significant effects or trends (all ps > .09), making it difficult to
67
interpret this interaction.
Interactions with Alexithymia
In this section, we test the hypothesis that higher levels of alexithymia attenuate faster
processing for negative and/or positive words. We tested the effect of alexithymia by adding
main and interactive effects for the total alexithymia score to the previous models. The total
alexithymia score for the participants ranged from 28-75 (M = 43, SD = 11.26). The total score
on the TAS-20 can range from 20-100, which indicates that low, moderate and high levels of
alexithymia are represented in our data. We restrict our interpretation to the effects that involve
alexithymia to avoid reproducing the analysis of the previous sections. For gaze durations (see
Table 5), there was no main effect for alexithymia (p > .77). There was a significant interaction
between valence and alexithymia for positive words (b = 0.04, SE = 0.02, p < .05), indicating
that the faster early-stage processing for positive compared to neutral words that we described
above was gradually attenuated as alexithymia increased (i.e., processing for positive words
became slower as alexithymia increased). This effect did not reach significance for negative
words. However, a model with negative valence as the baseline showed that there was no
difference between negative and positive words (all ps > .17). The partial effects for the valence
by alexithymia interaction are plotted in Figure 2.
68
Table 5. Interaction between Valence (Negative [Neg], Neutral [baseline], Positive [Pos]), Frequency (Freq: high, low [baseline]),
Concreteness (Conc: continuous), and Alexithymia (Alex: continuous) With Covariate for Length (continuous)
Gaze duration
Second pass time
Fixation probability
Regression probability
b
SE
p
b
SE
p
b
SE
p
b
SE
p
(Intercept)
5.59 0.03 0.0001*** 5.65 0.04 0.0001*** 1.78 0.13 0.0000*** -1.86 0.19 0.0000***
Neg
-0.04 0.02 0.0808
-0.12 0.05 0.0180*
0.13 0.13 0.3391
-0.05 0.16 0.7598
Pos
-0.07 0.02 0.0036** -0.13 0.05 0.0140*
0.09 0.16 0.5889
-0.13 0.17 0.4312
Freq
-0.08 0.03 0.0036** -0.03 0.06 0.6524
-0.07 0.13 0.5836
0.11 0.2 0.5847
Conc
-0.03 0.02 0.0364*
-0.04 0.03 0.2588
0.09 0.09 0.3422
0.08 0.12 0.5071
Alex
-0.01 0.02 0.779
-0.03 0.04 0.5138
0.1
0.14 0.4438
-0.05 0.16 0.7479
Length
0.07 0.01 0.0001*** 0.07 0.02 0.0004*** 1.01 0.06 0.0000*** 0.11 0.08 0.1436
Neg by Freq
0.05 0.04 0.1834
-0.04 0.08 0.5386
-0.19 0.18 0.2939
-0.05 0.24 0.8475
Pos by Freq
0.03 0.03 0.313
0.01 0.07 0.9492
-0.1 0.19 0.6121
-0.08 0.24 0.7371
Neg by Conc
0.06 0.02 0.0068** 0.04 0.05 0.3982
-0.09 0.13 0.5005
-0.12 0.16 0.4455
Pos by Conc
0.06 0.02 0.0068** 0.01 0.05 0.7698
0.07 0.15 0.6693
-0.28 0.16 0.077
Freq by Conc
0.06 0.02 0.0154*
0.04 0.05 0.3992
-0.1 0.12 0.4161
-0.02 0.18 0.9313
Neg by Alex
0.02 0.02 0.342
-0.01 0.05 0.8236
-0.24 0.13 0.077
0.03 0.13 0.8394
Pos by Alex
0.04 0.02 0.0306*
0.09 0.05 0.0744
-0.15 0.16 0.3581
0.11 0.14 0.4517
Freq by Alex
0.02 0.02 0.3022
0.06 0.05 0.2846
-0.06 0.13 0.6298
0.19 0.14 0.1657
Conc by Alex
0.02 0.01 0.065
0
0.03 0.9776
0.04 0.09 0.7093
0.08 0.09 0.3451
Neg by Freq by Conc
-0.1 0.04 0.0040** -0.05 0.08 0.5266
0.28 0.18 0.1283
-0.05 0.24 0.8289
Pos by Freq by Conc
-0.08 0.03 0.0112*
0
0.07 0.998
-0.02 0.19 0.9158
0.5
0.22 0.0247*
Neg by Freq by Alex
-0.02 0.02 0.3664
-0.08 0.07 0.3074
0.2
0.18 0.2788
-0.09 0.2 0.6427
Pos by Freq by Alex
-0.03 0.02 0.2012
-0.18 0.07 0.0130*
0.11 0.19 0.5624
-0.45 0.2 0.0254*
Neg by Conc by Alex
-0.02 0.02 0.2334
0.01 0.05 0.9052
-0.07 0.13 0.5831
-0.18 0.13 0.1822
Pos by Conc by Alex
0
0.02 0.8464
-0.02 0.05 0.5902
-0.02 0.16 0.9141
0.05 0.13 0.6869
Freq by Conc by Alex
-0.04 0.02 0.0220*
-0.04 0.05 0.4966
-0.1 0.13 0.4383
0.07 0.14 0.6038
Neg by Freq by Conc by Alex
0.05 0.03 0.075
0.04 0.08 0.6612
0.17 0.18 0.3543
-0.15 0.21 0.4826
Pos by Freq by Conc by Alex
0.02 0.02 0.4878
0.08 0.07 0.2458
0.08 0.19 0.6742
-0.24 0.2 0.2269
* p < 0.05, ** p < 0.01, *** p < 0.001
69
Figure 2. Partial effects for interaction between valence, frequency, and alexithymia showing
reduced processing benefit for low frequency positive words when alexithymia increases, which
eliminates advantage compared to neutral words
There was also a significant interaction between frequency, concreteness, and alexithymia
for gaze durations (b = -0.04, SE = 0.02, p < .05). Thus, we fit separate models to the data for
low and high frequency items to test the hypothesis that high alexithymia attenuates faster
processing for concrete compared to abstract words. These analyses revealed that when
alexithymia was low, processing was faster for words higher in concreteness and low in
frequency. However, this processing advantage for concrete words compared to abstract words
was attenuated as alexithymia increased. This effect did not interact with emotional valence,
indicating that it occurred for negative, neutral, and positive words (all ps > .24). There were no
significant effects when words were high in frequency (all ps > .12). The partial effects for the
interaction between frequency, concreteness, and alexithymia are plotted in Figure 3.
70
Figure 3. Partial effects for interaction between frequency, concreteness, and alexithymia
showing reduced processing benefit for neutral low frequency concrete words when alexithymia
increases, which reduces advantage compared to abstract neutral words
For late-stage processing measured by second pass time, there was a three-way
interaction between valence, frequency, and alexithymia for positive words (b = -0.18, SE = 0.07,
p < .05). We fit separate models to the low and high frequency items to test the hypothesis that
alexithymia attenuates faster processing for positive words. This analysis indicated that the
processing advantage for positive words relative to neutral words was attenuated as alexithymia
increased (i.e., late-stage processing for positive words became slower as alexithymia increased)
when words were low in frequency. However, there was no alexithymia effect for high frequency
words (all ps > .1). A model with negative valence as the baseline indicated that processing for
positive words also became slower relative to negative words as alexithymia increased (b = 0.1,
SE = 0.05, p < .05). Thus, alexithymia attenuated faster late-stage processing for low frequency
71
positive words.
There were no significant alexithymia effects on fixation probability, and a single
significant interaction effect for regression probability between valence, frequency, and
alexithymia for positive words (b = -0.45, SE = 0.2, p < .05) (see Table 5). Separate models fit to
the low and high frequency data showed that the interaction for regression probability reflected a
smaller probability of making regressions to high frequency positive words as alexithymia
increased compared to high frequency neutral words (b = -0.35, SE = 0.14, p < .05), and no
alexithymia effect for low frequency words (all ps > .17).
Discussion
We used eye movement measures of natural reading to investigate questions about
emotional word processing that emerge from an embodied view of language (Barsalou et al.,
2008; Juhasz et al., 2011; Kousta et al., 2011; Vigliocco et al., 2009). According to this view, the
effect of embodied information grounded in emotional and sensorimotor experiences should be
attenuated when linguistic information is sufficient for a response. Thus, assuming that word
frequency reflects a linguistic contribution to representation, the effect of emotional and
sensorimotor information should be maximal when frequency is low. Moreover, emotional
benefits should be maximal when words do not already enjoy a processing advantage from
associations with sensorimotor information (and vice versa). Thus, we investigated the following
three questions: (1) Are emotional words prioritized when people read naturally for
comprehension? (2) Are emotional processing advantages modulated by linguistic and
sensorimotor information? and (3) Do individual differences in emotional or sensorimotor
processing alter the basic pattern of emotional processing benefits?
Regarding the first two questions, our results for early-stage processing show that when
words are low in frequency, fixation durations during natural reading are shorter for negative and
positive words, which do not differ from each other, compared to neutral words. Moreover, we
72
also demonstrate that emotional words are maximally faster compared to neutral words when
words are also low in concreteness (i.e., abstract) in addition to being low in frequency. Thus,
these results suggest that emotional processing benefits emerge when words do not already enjoy
a processing advantage from rich linguistic or sensorimotor information.
Our results extend the embodied work on word representation that inspired our
hypotheses (Kousta et al., 2011; Vigliocco et al., 2009). Vigliocco and colleagues proposed that
abstract and concrete words are represented mostly by emotional and sensorimotor information,
respectively, in addition to being represented by linguistic information. We extend their work by
showing that embodied sources of information do not produce behavioral effects when rich
linguistic information is sufficient for generating a response (i.e., saccades in our natural reading
task), consistent with previous work indicating that responses can occasionally be generated
using only linguistic information (Barsalou et al., 2008; see also Juhasz et al., 2011). Note that
we are suggesting that linguistic information is sufficient for generating saccades away from
words, not meaningful comprehension (we further discuss this below while addressing models of
eye movements during reading).
We also extend the work of Vigliocco and colleagues by showing that when words have
rich associations with emotional and sensorimotor information, word processing will benefit
from either emotional information or sensorimotor information, but not both. Specifically, we
found that early-stage processing for concrete words rich in sensorimotor information did not
benefit further from rich emotional information (i.e., negative or positive valence). However, this
interaction can also be characterized the other way around: Processing for emotionally charged
negative and positive words did not benefit further from rich sensorimotor information. Thus, it
is unclear which representational content, emotional or sensorimotor, underlies processing
benefits for the particular subset of words that are both emotionally charged and concrete. Very
few studies have examined how sensorimotor information modulates emotional processing
73
benefits (or vice versa). The findings of a previous ERP study that manipulated emotional
valence and concreteness suggest that concrete negative words differ from neutral and positive
words as a function of mental imagery (Kanske & Kotz, 2007). However, comparison between
that study and the present one is complicated by Kanske and Kotz’s use of a hemifield paradigm
and the lack of a word frequency manipulation. Thus, more work is necessary for elucidating the
processing dynamics of emotionally charged concrete words.
The results of the present study can also be integrated with computational models of
reading (e.g., Reichle, Rayner, & Pollatsek, 2003). According to such models, the oculomotor
system occasionally begins programming a saccade out of a word after an initial analysis
indicates that word recognition is imminent. On such occasions, the word continues to be
processed (more fully) while the oculomotor system programs a saccade to the following word.
The type of information processed during the initial analysis has not yet been fully elucidated.
For example, either semantic or linguistic word-form information, or both, may be used to
initiate saccade programming (Reichle, Pollatsek, & Rayner, 2006).
We propose that linguistic information and embodied emotional and sensorimotor
information are all activated by every word during reading. When a word is high in frequency,
processing time will not be influenced by emotional and sensorimotor information because
linguistic information is sufficient to start programming a saccade out of the word before
embodied information is fully activated (though that embodied information is still retrieved with
additional processing while the saccade is being programmed). However, processing time for
harder-to-recognize low frequency words is influenced by embodied information because
saccade programming cannot begin until embodied information helps recognize the word.
Moreover, emotional and sensorimotor information interact when embodied sources of
information become necessary to start saccade programming. Specifically, for negative and
positive low frequency words, rich emotional information is sufficient to start saccade
74
programming. Thus, sensorimotor information will not provide any additional processing
advantage. Conversely, rich sensorimotor information associated with concrete words maximally
affects saccade programming when processing is not already enhanced by emotional charge (i.e.,
neutral words). This view of reading is compatible with the Language and Situated Symbol
theory (Barsalou et al., 2008), which proposes that linguistic information can be sufficient for
producing a response under certain conditions (i.e., high frequency words), but that embodied
information becomes necessary to produce a response in other cases (i.e., low frequency words).
Note that the interaction between linguistic and embodied information applies to early stages of
processing when the brain attempts to quickly determine if word recognition is imminent before
beginning saccade programming to the next word. During later stages, positive and negative
words are processed faster compared to neutral words irrespective of frequency and
concreteness.
Interestingly, emotional and sensorimotor information produced similar behavioral effects
during early-stage processing but dissimilar effects during late-stage processing. Specifically,
both emotional charge and concreteness produced faster early-stage processing but only
emotional charge produced faster late-stage processing. It is possible that sensorimotor
information is more heavily weighted during early compared to late processing because the
physical properties of referents may be primarily used for word-level processing, which should
maximally influence early reading measures. In contrast, emotional information may be used for
word-level and sentence-level processing because it indexes both the nature of referents and their
contextual value, and the latter should maximally influence late reading measures. Future work
will need to confirm these speculations.
Our findings are also important in light of past studies of emotional word processing,
which have been contradictory. Some studies observed faster responses for negative and positive
words compared to neutral words (Kousta et al., 2009; Long & Titone, 2007, Holt et al., 2006),
75
and others observed faster responses for negative words only when words were low in frequency
(Kuchinke et al., 2007; Méndez-Bértolo et al., 2011; Scott et al., 2009, 2012), but faster
responses for positive words irrespective of frequency (Kuchinke et al., 2007; Scott et al., 2009,
2012). Thus, our findings and the results of Scott et al. (2012), who also used an eye tracking
sentence reading paradigm, diverge on a single result: Positive high frequency words were not
faster compared to neutral high frequency words in our study, but they were faster than neutral
words in Scott et al. (2012).
The divergence on high frequency positive words may reflect cross-study variability in
arousal. For example, high frequency negative and positive words in Scott et al. (2012) had
higher arousal values (M = 6.6, SD = 0.5; M = 6.4, SD = 0.6, respectively) than in our study (M =
5.25, SD = 0.65; M =5.54, SD = 0.88, respectively). We confirmed that the pattern of effects in
our study was driven by valence and not arousal by adding a covariate for arousal, which did not
change the effect of valence or any of the interactions. We also tested whether arousal produced
interactive effects with valence, frequency, or concreteness. However, including an interaction
with arousal was not statistically justified. Thus, faster processing for high frequency positive
words may depend on a combination of positive valence and arousal levels higher than in our
study. Variability in arousal can also account for differences between our findings and the lexical
decision results of Scott et al. (2009), and potentially the lexical decision results of Kuchinke et
al. (2007), though we cannot compare the arousal values of our stimuli with those of Kuchinke et
al. because they did not report mean arousal values for their materials.
According to the LASS theory (Barsalou et al., 2008), embodied information should have
no functional effect when rich linguistic information is sufficient for a response (i.e., when words
are high in frequency), which is at odds with Scott et al. (2009, 2012) who found faster
processing for high frequency positive words that were high in arousal. One assumption in the
LASS theory that is subject to debate is that linguistic information/lexical representations
76
provide only superficial non-semantic information. Although our findings are agnostic with
respect to that debate, they do suggest, when contrasted with the results of Scott et al. (2009,
2012), that a combination of high arousal and positive valence may enhance linguistic
information, resulting in faster sensory processing for high frequency positive words. This
possibility is not necessarily inconsistent with the LASS theory, as linguistic information may
simply be enhanced (e.g., via higher baseline activation) by high arousal, rather than encoding
information that can mediate meaningful comprehension.
Thus, it would be fruitful to manipulate the arousal and frequency of positive words in a
future study, which could potentially show that a different mechanism underlies emotional
processing benefits for low and high frequency positive words. For example, processing
advantages for low frequency positive words may reflect embodied simulations predicted by
valence but not arousal, but processing advantages for high frequency positive words may reflect
enhanced sensory processing predicted by arousal but not valence. Crucially, this future study
would have to use words higher in arousal than in the present study. The idea that emotion can
enhance lexical representations that mediate sensory processing has been proposed before
(Kousta et al., 2009; Scott et al., 2009, 2012), as has the potentially critical role played by
emotional arousal in enhancing the sensory processing of words (Anderson, 2005).
Our findings also have important implications for work on how emotional words are
processed in sentences, which, to our knowledge, consists only of three previous studies. Two
event-related potential (ERP) studies indicate that compared to neutral words, the late positive
complex is enhanced for negative words (Bayer, Sommer, & Schacht, 2011; Holt et al., 2009)
and the N400 is enhanced for negative and positive words (Holt et al., 2009). Thus, these studies
suggest that emotional target words in emotionally neutral sentential contexts recruit more
resources compared to neutral targets, which should result in slower processing. However, these
studies did not manipulate word frequency, which is critical according to a growing number of
77
studies as well as the present study. Indeed, as we have already mentioned above, our results are
largely consistent with the only other eye tracking study of sentence reading that manipulated
both emotional valence and word frequency (Scott et al., 2012). Although Scott et al. (2012)
produced a single result that is at odds with our findings (which, as we have discussed above,
may reflect differences in arousal), they also observed faster processing for low frequency
emotional words compared to neutral words for eye movement measures of early-stage
processing. For eye movement measures of late processing (i.e., total reading time), Scott et al.
(2012) stated that the interaction between frequency and emotional valence was attenuated,
though they did not present a statistical analysis for total reading time. Thus, our study supports
work showing that emotional information produces faster processing for low frequency
emotional words during early stages of processing, and faster processing for all words during
later stages of processing.
The embodied framework that guided our research suggests that individual differences
that influence emotional and sensorimotor processing should modulate emotional and
concreteness benefits, respectively, which leads to the third main question of the study regarding
individual differences in alexithymia. We found that alexithymia attenuated the emotional benefit
for positive words compared to neutral words during early- and late-stage processing. Thus, our
results demonstrate that emotional processing deficits in single word paradigms predicted by
alexithymia generalize to natural reading (Mueller et al., 2006; Luminet et al., 2006; Goerlich et
al., 2011; respectively). However, contrary to the results of these studies, alexithymia did not
predict processing deficits for negative words. It is possible that alexithymia affects negative
words when single word paradigms direct neurocognitive resources toward single words, but
does not affect processing when a reading task directs resources more towards building up
meaning for the sentence as a whole. Alexithymia effects on positive words may be more robust
by comparison, irrespective of whether resources are mostly allocated to single words or
78
sentence meanings. Some evidence for weaker alexithymia effects on negative words can be
found in the cited studies. For example, alexithymia effects on ERPs were topographically more
widespread for positive compared to negative valence during affective mismatches (Goerlich et
al., 2011). In addition, the effect of alexithymia on negative words was conditional on level of
positive and negative state affect, optimism, and depression (Luminet et al., 2006). However, the
alexithymia effect on positive words in the same study was significant with and without these
covariates.
Interestingly, we also found that alexithymia attenuated the early-stage processing
advantage for concrete words compared to abstract words. One account of the concreteness
advantage posits that concrete words are processed faster compared to abstract words because
concrete words have more associations with sensorimotor information (e.g., visual information in
Paivio’s dual-coding theory, 1971). Previous work has shown that people high on alexithymia
show hypersensitivity to sensorimotor information (Nyklicek & Vingerhoets, 2000; Schafer et
al., 2007; Sivik, 1993). Thus, alexithymia may attenuate processing differences between concrete
and abstract words because sensorimotor information is amplified and overrepresented across all
words, reducing the degree to which concrete and abstract words are differentially associated
with sensorimotor information.
Taken together, the findings of this study demonstrate for the first time that word
processing during natural reading is jointly influenced by linguistic, emotional, and sensorimotor
information, producing interactive effects during early stages of processing. We also show for the
first time that individual differences in eye movement measures of natural reading reflect the
nature of individuals’ emotional and sensorimotor experiences. Emotional information has
traditionally been neglected in cognitive theories of representation. Barsalou (2008) argues that
even embodied approaches are often misconstrued as arguing that conceptual content is
grounded only in perception of the external world. Our work thus shows that emotional
79
information plays a fundamental role in word processing, consistent with a small body of work
which shows that emotional information plays a central role in representing concepts,
particularly abstract ones (Barsalou & Wiemer-Hastings, 2005; Kousta et al., 2011; Vigliocco et
al., 2009). These studies suggest that emotion grounds abstract concepts (which have been a
classic problem for embodied theories of representation) in internal states, which may bootstrap
the acquisition of abstract semantics during development in infancy (Kousta et al., 2011).
80
PREFACE FOR CHAPTER 3
The study presented in Chapter 2 examined how emotional, sensorimotor, and linguistic
contributions to representation (indexed by emotional valence, concreteness, and word
frequency, respectively) jointly shape word processing in native speakers of English that read
sentences in their L1. The research questions focused on early stages of word processing, which
were measured with early eye-movement measures of reading. The study found that people read
negative and positive words faster than neutral words, and that this emotional advantage is
maximal for words low in concreteness (i.e., abstract words) and low in word frequency. The
study also found that people read concrete words faster than abstract words among words that are
emotionally neutral and low in frequency. Thus, rich emotional information functionally
modulates early eye-movement measures when sensorimotor and linguistic contributions to
representation are insufficient for word processing. Similarly, rich sensorimotor information
functionally modulates early eye-movement measures when emotional and linguistic
contributions to representation are insufficient for word processing. Thus, the findings are
consistent with the idea that any given contribution to representation will functionally modulate
behavior only when other contributions to representation (whether embodied or linguistic) are
insufficient for generating a response (Barsalou et al., 2008).
However, an open question concerns whether emotion plays the same role in bilinguals
processing words in their L2. Sociolinguistic work on how bilinguals deploy their languages
across different social contexts shows that bilinguals prefer to use their L1 over their L2 for
emotional social exchanges (Dewaele, 2004; Pavlenko, 2005). Thus, there may be few
opportunities for L2 words to co-occur with emotional experiences, which would leave L2 words
with emotionally impoverished or disembodied semantic representations (Pavlenko, 2012) given
that semantic representations are embodied in experiences that co-occur with word use
(Vigliocco et al., 2009; Zwaan, 2008). Moreover, several lines of work suggest that negative
81
words in particular may be at risk of emotional disembodiment (Boucher & Osgood, 1969;
Conrad et al., 2011; Matsumoto et al., 2005).
Thus, the study in Chapter 3 tests whether negative words, but not positive words, are
emotionally disembodied in bilinguals (native speakers of French) reading English sentences in
their L2. The bilinguals read the same sentences as the native speakers of English in Chapter 2.
Thus, Chapter 3 can compare how word processing is jointly influenced by emotional,
sensorimotor, and linguistic contributions to word representation in L2 processing vs. L1
processing.
82
CHAPTER 3:
The Embodiment of Emotional Words in a Second Language:
An Eye-Movement Study.
(Sheikh & Titone, 2015, Cognition And Emotion)
83
Abstract
The hypothesis that word representations are emotionally impoverished in a second
language (L2) has variable support. However, this hypothesis has only been tested using tasks
that present words in isolation, or that require laboratory-specific decisions. Here, we recorded
eye movements for 34 bilinguals who read sentences in their L2 with no goal other than
comprehension, and compared them to 43 first language readers taken from our prior study
(Sheikh & Titone, 2013). Positive words were read more quickly than neutral words in the L2
across first-pass reading time measures. However, this emotional advantage was absent for
negative words for the earliest measures. Moreover, negative words but not positive words were
influenced by concreteness, frequency, and L2 proficiency in a manner similar to neutral words.
Taken together, the findings suggest that only negative words are at risk of emotional
disembodiment during L2 reading, perhaps because a positivity bias in L2 experiences ensures
that positive words are emotionally grounded.
84
Introduction
Emotion is thought to play a foundational role in grounding semantic representations
during first language (L1) processing (Kousta, Vigliocco, Vinson, Andrews, & Del Campo,
2011). However, the role of emotion in grounding second language (L2) semantic
representations remains an open question: Some studies find that bilinguals process emotional
and neutral words (e.g., sex vs. pin, respectively) differently in their L2, similar to what native
speakers do in their L1 (e.g., Sutton, Altarriba, Gianico, & Basnight-Brown, 2007), whereas
other studies do not (e.g., Degner, Doycheva, & Wentura, 2012). Such discrepancies may arise
for a variety of potential reasons, which include whether bilinguals process emotional words
differently as a function of L2 proficiency, emotional polarity, or comprehension demands. Here,
we test these alternatives, with the ultimate goal of determining whether an embodied theoretical
approach to language (e.g., Barsalou, 1999) can explain when bilinguals process L2 emotional
words like L1 emotional words. We first review the bilingual literature on L2 emotional word
processing, and then describe the embodied approach that comprises the theoretical basis of this
study.
One source of complexity regarding L2 emotional word processing concerns L2
proficiency. For example, bilinguals who are as proficient in their L2 as their L1 show intact
emotional word processing in emotional Stroop tasks (Eilola, Havelka, & Sharma, 2007; Sutton
et al., 2007). Conversely, bilinguals who are less proficient in their L2 than their L1 show
reduced or no emotional effects in other tasks. For example, bilinguals show reduced skin
conductance responses to L2 childhood reprimands and taboo words in emotion rating tasks
(Harris, Ayçiçeği, & Gleason, 2003; see also, Segalowitz, Trofimovich, Gatbonton, &
Sokolovskaya, 2008; Degner et al., 2012, for implicit affect association and affective priming
tasks, respectively). The fact that low L2 proficient bilinguals show impaired emotional
processing effects implies that bilinguals require additional, and likely direct experience with L2
85
words in order to treat them emotionally. It also implies that automatic cross-language activation
that routinely occurs during bilingual language processing, which is highly probable for low L2
proficient bilinguals (e.g., Thierry, G., & Wu, 2007; reviewed in Baum & Titone, 2014; Kroll,
Bobb, Misra, & Guo, 2008), is insufficient for leading to emotional word processing effects in
the L2.
Another source of complexity with respect to L2 emotional word processing are potential
effects of emotional polarity (i.e., negative vs. positive emotionality). Most work focuses
exclusively on negative words—thus, few have compared negative and positive words to each
other, while simultaneously taking into account the other linguistic ways that words vary (e.g.,
frequency). A recent exception is Conrad et al. (2011), who compared emotional and nonemotional words using event-related potentials (ERP) gathered during lexical decisions in two
groups of bilinguals differing in L2 proficiency. Here, more proficient bilinguals showed an
enhanced early posterior negativity and late positive complex in L1 for negative and positive
words vs. neutral words, with delayed but similar effects in the L2. Less proficient bilinguals,
however, showed ERP modulations in the L2 for positive but not negative words. Thus, negative
but not positive emotional words may be treated in an unemotional manner in the L2. In contrast
with this work, Degner et al. (2012) and Segalowitz et al. (2008) found impoverished emotional
word processing in an implicit affect association task and an affective priming task, respectively,
which did not interact with continuous measures of L2 proficiency. Interestingly, however,
Degner et al. found that bilinguals who reported frequently using their L2 processed emotional
words in their L2 more like they did in their L1, even after controlling for L2 proficiency.
Thus, an open question concerns the amount and kind of L2 experience required for
intact emotional word processing. One potentially useful approach for addressing this question
is to view emotional effects on word processing as part of a larger class of language embodiment
phenomena, which considers how real-world experiences ground semantic representations
86
generally. Consider the words apple and grief, for which people have very different bodily
experiences. Apples are tangible objects that may be grasped, eaten, or physically manipulated
in any number of ways (e.g., sliced, cooked, stewed, etc.), which are all typically classed as
sensorimotor experiences. Grief is a less tangible concept that may have indirect physical
manifestations that we typically class as emotional experiences (e.g., a lump in one’s throat). In
embodied approaches to language, words are grounded in whichever type of experience they
tend to co-occur with during language use (Zwaan, 2008). Vigliocco and colleagues recently
proposed that all words are represented by linguistic information (e.g., word associations),
emotional information, and sensorimotor information (Kousta et al., 2011). They further
stipulated that emotional and sensorimotor information primarily represent abstract and concrete
semantics, respectively.
With respect to L2 word processing in bilinguals, an embodied approach to language
forces us to consider more deeply the behavioral ecology of bilinguals (Green, 2011), and
whether this ecology would afford the embodiment of emotional words in one’s L2.
Accordingly, bilinguals' L2 words may be “disembodied” (Pavlenko, 2012) if they do not use
their L2 in social contexts that provide opportunities for co-occurrence with emotional
experiences. This idea is consistent with work on language choice. For example, bilinguals
curse to express anger using their L1 more than their L2 (Dewaele, 2004), and switch to their L1
for emotionally charged interactions with their romantic partners, even when their partners have
limited knowledge of the language (Pavlenko, 2005). Thus, bilinguals may experientially lock
their L2 words out of emotional contexts, thus rendering them less salient than L1 words during
ongoing language processing.
Bilinguals might also show differences between negative and positive words to the extent
that they experientially lock their L2 out of negative but not positive emotional contexts. For
example, Conrad et al. (2011) suggested that bilinguals showed no emotional effects for negative
87
words but not positive words because a positivity bias ensures that bilinguals use their L2 in
emotionally positive contexts. The notion of a positivity bias is supported by work on the
Pollyanna hypothesis showing that human communication is centered on emotionally positive
exchanges (Boucher & Osgood, 1969). It is also supported by work on emotion regulation
showing that positive emotion is more often up-regulated and less often down-regulated than
negative emotion (Matsumoto, Yoo, Hirayama, & Petrova, 2005), particularly with respect to
interactions with colleagues and strangers vs. family and friends. An embodied perspective also
forces us to consider differences between emotional and sensorimotor embodiment in
bilinguals—which are reflected in emotional and concreteness advantages, respectively (Kousta
et al., 2011). Accordingly, bilinguals may have difficulty grounding L2 words in emotional
experiences specifically but not sensorimotor experiences more generally. To test this idea, we
can capitalize on interactions found in the L1 literature between word emotionality and
concreteness, which shows that emotion is more likely to facilitate semantic categorization
accuracy for abstract but not concrete words (Newcombe, Campbell, Siakaluk, & Pexman,
2012). Similarly, in eye-movement paradigms, emotion facilitates first-pass reading times for
abstract words and not concrete words, and conversely, concreteness facilitates neutral words but
not emotional words (Sheikh & Titone, 2013). Thus, if bilinguals have disembodied negative L2
words, those negative words should be facilitated by concreteness, like neutral words, and not by
emotionality.
Differences among bilinguals in L2 proficiency may not have predicted emotional effects
in previous studies because proficiency does not reflect the kinds of contexts in which bilinguals
use their L2. However, L2 proficiency might modulate facilitation by concreteness to the extent
that sensorimotor referents of words are independent of context. For example, there is no
evidence suggesting that sensorimotor experiences vary as a function of social context, as has
been shown for emotional experiences (Matsumoto et al., 2005). Thus, L2 proficiency should
88
modulate concreteness advantages for negative and neutral words. Moreover, L2 proficiency
should modulate word frequency effects, which reflect how often words occur in language, to the
extent that frequency effects also do not depend on context-specific L2 experiences. Consistent
with this conjecture, differences in language use measures among bilinguals predict L2 and L1
frequency effects (e.g., Whitford & Titone, 2012) even though the usage measures do not reflect
the context of use. Bilinguals generally also show larger L2 than L1 frequency effects,
presumably because bilinguals experience words less frequently in their L2 (e.g., Whitford &
Titone, 2012). Thus, bilinguals should show less dependence on word frequency for modulating
embodiment effects, in contrast to native speakers (Juhasz, Yap, Dicke, Taylor, & Gullick, 2011;
Sheikh & Titone, 2013).
The Present Study
The work just reviewed leads to the following open questions about bilinguals reading in
their L2: (1) Do bilinguals show reduced or eliminated emotional facilitation for negative words,
but not positive words, relative to neutral words? (2) Do bilinguals show facilitation of negative
words by concreteness, like neutral words? (3) Do differences in L2 proficiency among
bilinguals predict facilitation by concreteness and frequency, but not emotionality? We address
these questions using eye-movement measures of natural sentence reading (Rayner, 2009). We
used semantically neutral sentences in which we embedded English target words that varied on
emotional valence, frequency, and concreteness.
French-English bilinguals (all French L1) read English sentences presented in their
entirety, with no goal other than comprehension. Since embodiment theories are based mostly on
data from single word paradigms, they do not help target particular reading time measures for the
disembodiment predictions. Thus, we analyzed all first-pass fixation time measures from first
fixation duration to go-past time to identify the point in time at which emotional disembodiment
manifests in bilinguals for negative words. To obtain more evidence that our bilinguals have
89
disembodied negative words but not positive words, we compared the bilinguals with native
English speakers from a previous study that used the same materials (Sheikh & Titone, 2013).
Method
Participants
We recruited 34 bilinguals (French L1, English L2, mean age = 24.97, SD = 5.16) at
McGill University that were less proficient in their L2 than their L1 and primarily used their L2
in formal environments rather than informal social environments (see supplementary materials in
Appendix B). We report below how we determined sample size, all data exclusions,
manipulations, and measures in the study.
Materials and Design
We used 156 target words grouped into 52 triplets, each consisting of a negative, neutral,
and positive word. Frequency and concreteness were manipulated, but balanced across
emotional categories (see Table 1). Length was longer for low compared to high-frequency
words, which was statistically controlled in all analyses. We created three sentences for each
triplet which were well-formed when combined with any of the triplet members. Differences in
sentential fit across emotional categories were avoided, and any potential effects were
statistically controlled (see Sheikh & Titone, 2013, where we used the same stimuli and describe
their development in depth). Targets and sentence frames were presented once in a given
experimental list and the combinations were counterbalanced across participants (see Table 2).
90
Table 1. Mean (Standard Deviations) For Lexical Attributes For Target Words And Their Latent Semantic Analysis
(LSA) Values Within Sentence Frames
Low frequency
High frequency
Negative
Neutral
Positive
Negative
Neutral
Positive
a
Log frequency
7.58 (1)
7.18 (1.27)
7.86 (1.03)
9.89 (0.77)
10.31 (0.91)
10.11 (0.91)
Valenceb
3.15 (0.58)
5.28 (0.51)
7.11 (0.42)
3.05 (0.85)
5.39 (0.54)
7.08 (0.45)
Arousalb
5.13 (0.8)
4.36 (0.87)
5.42 (1.07)
5.25 (0.65)
4.54 (0.73)
5.54 (0.88)
c
Concreteness
429.62 (106.9) 463.6 (138.2) 459.14 (125.8) 445.4 (120.85) 471.93 (116.24) 468.42 (129.05)
Length
6.31 (2.32)
7.2 (3.52)
7 (2.14)
5.9 (2.34)
4.81 (2.37)
5.39 (1.75)
LSA value
-0.005 (0.06)
0.013 (0.06) 0.005 (0.06)
0.004 (0.05)
0.009 (0.04)
0.010 (0.06)
a
The English Lexicon Project (Balota et al., 2007)
b
Kousta et al. (2009)
c
MRC Psycholinguistic Database (Coltheart, 1981)
Table 2. Target Words (Between Asterisks) Were Presented Within One Of Three Sentence Frames in a
Given List, Which Were Rotated Around Their Triplet’s Negative, Neutral, and Positive Target Words
Across Lists so That Targets Were Fully Crossed With Sentence Frames
Counterbalance
Sentence
List 1
The art teacher presented the *smoke* that the students were going to paint.
The news report mentioned the *hill* that was discussed at work.
The school paper reviewed the *drink* that had caused concern among students.
List 2
The art teacher presented the *drink* that the students were going to paint.
The news report mentioned the *smoke* that was discussed at work.
The school paper reviewed the *hill* that had caused concern among students.
List3
The art teacher presented the *hill* that the students were going to paint.
The news report mentioned the *drink* that was discussed at work.
The school paper reviewed the *smoke* that had caused concern among students.
Note. Targets were not presented with asterisks during the experiment.
91
A modified Language Experience and Proficiency Questionnaire (LEAP-Q; Marian,
Blumenfeld, & Kaushanskaya, 2007) was also administered. We used the LEAP-Q to verify that
all participants were native speakers of French and dominant in their native language, that
English was their L2, and to measure L2 proficiency using a 7-point scale that ranged from 1
(beginner) to 7 (near-native).
Apparatus and Procedure
Eye movements were recorded using an Eyelink 1000 that sampled eye position every
millisecond. The right eye was recorded but viewing was binocular. Sentences were displayed
on a 21-in. Viewsonic CRT monitor with 144 Hz refresh rate, 1024 × 768 resolution, yellow 10point Monaco font on black background, with three characters subtending approximately 1° of
visual angle. Yes or no questions followed 21% of trials to ensure participants read for
comprehension. The LEAP-Q was administered after the reading task.
Results
We tested our hypotheses using first-pass fixation duration measures, which reflect early
stages of lexical processing (rather than later measures that reflect the integration of word
meaning into the sentential context) (Rayner, 2009). We analyzed first fixation duration (FFD;
the duration of the first fixation on a word), single fixation duration (SFD; fixation time in cases
where the word was fixated exactly once), gaze duration (GD; the sum of the durations of all
fixations made during the first pass, before the eyes left the target region), and go-past time
(GPT; also known as regression path duration; the sum of the durations of all fixations on the
word from the point when the word is first fixated up until the eyes move past the word to the
right). We also analyzed the probability of fixating and regressing to a target, which are
presented as supplementary material (see Appendix B).
The eye-movement measures were analyzed in R (R Development Core Team, 2010)
with LMMs (and generalized LMMs for the binary data) using the lme4 package (Bates et al.,
92
2009). To test our hypotheses regarding negative and positive valence, we specified predictors
for valence (negative vs. neutral vs. positive), frequency (continuous), concreteness (continuous),
L2 proficiency (continuous), and their interactions in the fixed-effect structure. The model
baseline was set to neutral valence because our hypotheses specifically concern differences
between negative and positive words vs. neutral words. All continuous predictors were
standardized. We also included a covariate for word length (continuous) in all analyses.
We maximized the random-effect structure to the extent possible (Barr, Levy, Scheepers,
& Tily, 2013). We included random intercepts for participant (subject), word, and triplet; and bysubject random slopes for valence (including the slope-intercept correlation) and frequency
(including the slope-intercept correlation for SFD, but not the other measures).
We also used likelihood ratio tests to confirm that the highest-order significant effect in
each model was justified to ensure that the data were not overfitted, though the full models are
presented in the tables below for the sake of comparison. The models were fit to log-transformed
observations to meet model assumptions, and nonlog predicted values were calculated for
plotting partial effects. We report the estimated coefficient (b), standard error (SE), and t value.
An absolute value of t ≥ 1.96 falls within the 95% confidence interval and is significant. For
convenience, we calculated p values using an upper bound for the degrees of freedom (number of
observations – the number of fixed effects in the model), which should not lead to anticonservative values according to Baayen (2008), since we have well over the minimum 100
observations that he indicates should suffice for calculating p values in this manner. As seen
below, there was no discrepancy between these p values and the |t| ≥ 1.96 criterion.
Means and standard deviations for the eye-movement measures are presented in Table 3.
Mean accuracy on the comprehension questions was 81.83% (SD = 4.85). Twenty-six percent of
the trials were discarded because of track losses, lack of fixation on the target words, blinks, or
when observations were shorter than 80 ms. Visually identified outliers were eliminated from
93
the remaining trials (FFD > 800 [0.57%], SFD > 800 [0.71%], GD > 1500 [0.22%], GPT > 1500
[0.32%]).
Table 3. Means (Standard Deviations) For Eye-Movement Measures for Target Words
Low frequency
Low concreteness
High concreteness
Measures
Negative
Neutral
Positive Negative
Neutral
Positive
First fixation
280 (107) 279 (105) 272 (102) 285 (116) 273 (100) 279 (107)
Single fixation
293 (113) 310 (118) 282 (103) 294 (115) 280 (101) 279 (104)
Gaze duration
378 (224) 435 (247) 371 (213) 362 (198) 372 (231) 330 (155)
Go-past time
418 (243) 485 (271) 397 (224) 374 (202) 394 (258) 342 (169)
Fixation probability
0.91
0.92
0.97
0.87
0.88
0.89
Regression probability 0.15
0.18
0.16
0.15
0.16
0.09
High frequency
Low concreteness
High concreteness
Negative
Neutral
Positive Negative
Neutral
Positive
First fixation
272 (108) 276 (108) 255 (89) 269 (93) 279 (109) 255 (88)
Single fixation
275 (106) 284 (114) 261 (92) 275 (97) 287 (112) 257 (88)
Gaze duration
321 (155) 323 (147) 297 (137) 309 (133) 306 (133) 284 (122)
Go-past time
333 (166) 339 (160) 311 (148) 319 (146) 327 (168) 297 (136)
Fixation probability
0.89
0.82
0.82
0.83
0.78
0.81
Regression probability 0.13
0.14
0.15
0.15
0.16
0.14
Note. Frequency and concreteness were continuous variables in the analyses.
Setting aside the effect of L2 proficiency, we can see in the LMM outputs (Table 4) that
negative and neutral words did not differ for FFD, SFD, or GD. However, the GPT model
showed a single main effect for negative valence, suggesting that negative words (with sufficient
processing time) eventually became faster than neutral words. In contrast to negative words,
significant main effects for positive valence across all measures indicate that positive words were
faster than neutral words. Moreover, in the FFD data, there was an unexpected two-way
interaction between positive valence and frequency indicating that the emotional advantage for
positive words was even larger for high-frequency words compared to low-frequency words.
Additional models with negative valence as the baseline (as opposed to neutral described above)
showed that positive words were faster than negative words across all measures (ps < 0.05).
94
Thus, the emotional advantage is reduced for negative words but not positive words, and
emotional advantages occur for both high-frequency and low-frequency words.
95
Table 4. Test Of Interaction Between Valence, Frequency, Concreteness, And Proficiency; Showing Emotional Advantage For Positive
Valence Across All Measures; Which Is Absent For Negative Valence except for Go-past time; And A Frequency × Concreteness ×
Proficiency Interaction That Is Different For Negative And Neutral Words Vs. Positive Words
First fixation
Single fixation
Gaze duration
Go-past time
Fixed effect
b
SE
T
b
SE
t
b
SE
t
b
SE
t
(Intercept)
5.56 0.03 185.02*** 5.61 0.03 172.27*** 5.76 0.04 158.94*** 5.82 0.04 153.52***
Neg
0
0.02 -0.05
-0.01 0.02 -0.48
-0.04 0.02 -1.83
-0.05 0.02 -2.4*
Pos
-0.03 0.01 -2.22*
-0.06 0.02 -3.63***
-0.09 0.02 -4.38***
-0.11 0.02 -5.1***
Freq
-0.01 0.01 -0.78
-0.02 0.01 -1.23
-0.06 0.02 -3.2**
-0.05 0.02 -2.81**
Conc
-0.01 0.01 -0.6
-0.01 0.01 -0.49
0.01 0.02 0.37
0.01 0.02 0.8
Prof
-0.02 0.03 -0.73
-0.03 0.03 -0.95
-0.07 0.03 -2.06*
-0.06 0.04 -1.82
Len
-0.01 0.01 -0.74
0.02 0.01 1.53
0.1
0.01 6.73***
0.14 0.02 9.12***
Neg × Freq
-0.01 0.02 -0.38
-0.01 0.02 -0.53
0
0.02 0.04
-0.03 0.02 -1.09
Pos × Freq
-0.03 0.01 -2.15*
-0.02 0.02 -1.14
-0.01 0.02 -0.37
0
0.02 0.04
Neg × Conc
0
0.01 -0.24
0
0.02 -0.22
-0.01 0.02 -0.57
-0.02 0.02 -0.81
Pos × Conc
0.02 0.01 1.58
0.03 0.02 1.46
0.02 0.02 1.19
0.02 0.02 0.77
Freq × Conc
-0.01 0.01 -0.85
0
0.01 -0.17
-0.01 0.01 -0.95
0
0.01 -0.2
Neg × Prof
0.01 0.01 0.49
-0.01 0.02 -0.31
-0.01 0.02 -0.34
-0.01 0.02 -0.71
Pos × Prof
0.02 0.01 1.4
0.02 0.02 0.98
0
0.02 0.21
-0.01 0.02 -0.49
Freq × Prof
0
0.01 0.07
0.01 0.01 0.65
0.04 0.01 2.63**
0.03 0.02 2.02*
Conc × Prof
-0.01 0.01 -0.72
-0.01 0.01 -0.52
0
0.01 -0.24
-0.01 0.01 -0.55
Neg × Freq × Conc
0.01 0.02 0.48
0
0.02 0.09
0.01 0.02 0.54
0
0.03 0.07
Pos × Freq × Conc
-0.01 0.01 -0.85
-0.01 0.02 -0.75
0
0.02 -0.19
-0.01 0.02 -0.45
Neg × Freq × Prof
-0.02 0.01 -1.18
-0.03 0.02 -1.39
-0.02 0.02 -0.83
-0.01 0.02 -0.74
Pos × Freq × Prof
0
0.01 0.16
0
0.02 -0.19
-0.03 0.02 -1.96
-0.02 0.02 -1.32
Neg × Conc × Prof
0.01 0.01 1.08
0.02 0.02 1.04
0.02 0.02 0.91
0.03 0.02 1.45
Pos × Conc × Prof
-0.01 0.01 -0.41
0
0.02 0.05
0
0.02 -0.12
0
0.02 -0.22
Freq × Conc × Prof
0.01 0.01 0.76
0.01 0.01 0.52
0.03 0.01 2.7**
0.03 0.01 2.95**
Neg × Freq × Conc × Prof 0.01 0.01 0.74
0.01 0.02 0.74
-0.03 0.02 -1.41
-0.03 0.02 -1.59
Pos × Freq × Conc × Prof
-0.01 0.01 -0.91
-0.02 0.02 -1
-0.04 0.02 -2.21*
-0.04 0.02 -2.6**
Note. Neg = negative; Pos = positive; Freq = frequency; Conc = concreteness; Prof = second language proficiency
* p < 0.050, ** p < 0.010, *** p < 0.001
96
Regarding the effect of L2 proficiency, there were no interactions for FFD or SFD. However,
there were L2 proficiency interactions for GD and GPT. In both cases, this manifested as a four-way
interaction between positive valence, frequency, concreteness, and proficiency. L2 proficiency never
interacted with negative valence. Thus, the effect of proficiency on GD and GPT was identical for
negative and neutral words, which manifested in both cases as a three-way interaction between
frequency, concreteness, and proficiency. We visualized the pattern of effects from the model for GD
using the coefficients for these significant interactions in a partial effects plot (see Figure 1).
Figure 1. Frequency, concreteness, and proficiency produced identical effects for negative and neutral
words, which differed from positive words.
97
We fit separate models to the GD data, split by median concreteness, to follow up on the fourway interaction in Figure 1, and to test our hypothesis that L2 proficiency predicts the concreteness
advantage but not the emotional advantage. For words low in concreteness (i.e., abstract words), there
was a main effect for positive valence (b = -0.1, SE = 0.03, t = -3.56, p < 0.001) and no interactions,
which indicates that for abstract words, processing was faster for positive compared to neutral words
irrespective of frequency and proficiency. This emotional advantage for abstract positive words can be
seen in all of the panels in Figure 1 at low concreteness values.
For words high in concreteness (i.e., concrete words), there was an interaction between positive
valence, frequency, and proficiency (b = -0.08, SE = 0.03, t = -2.86, p < 0.01). This interaction
indicates that frequency and proficiency produced identical effects for concrete negative and neutral
words, but different effects for positive words. Specifically, concreteness facilitated processing for
negative and neutral low-frequency words, but only at high levels of proficiency. And as words
became more concrete, this reduced the emotional advantage for positive words. This pattern can be
seen in Figure 1 in the bottom right panel at high concreteness values. We also found this pattern of
effects in separate models fit to the GPT data split by concreteness. Thus, the results show that
concreteness advantages, but not emotional advantages, depend on proficiency, and that concreteness
produces identical effects for negative and neutral words, but different effects for positive words.
L2 Readers vs. L1 Readers
Next, we test differences between the bilinguals reading in their L2 (L2 readers) in this study
and 43 native speakers of English1 (L1 readers) that read the same materials in a previous study
(Sheikh & Titone, 2013). To compare L2 vs. L1 readers, we tested whether language group interacts
with valence, frequency, and concreteness. For a complete exposition of the L1 data, see Sheikh and
Titone (2013). The most relevant finding in the previous study is that the concreteness advantage was
limited to low-frequency neutral words in L1 readers, producing three-way interactions between
valence, frequency, and concreteness for negative and positive words for the first-pass fixation time
98
measures. Thus, if emotional disembodiment is specific to negative words in L2 readers, a comparison
with L1 readers should produce four-way interactions between language group, valence, frequency, and
concreteness for negative words, but not positive words. As expected, this four-way interaction was
significant for negative words for FFD (b = 0.04, SE = 0.02, t = 2.09, p < 0.05), SFD (b = 0.04, SE =
0.02, t = 1.93, p = 0.05), GD (b = 0.09, SE = 0.02, t = 4.01, p < 0.001), and GPT (b = 0.06, SE = 0.02, t
= 2.53, p < 0.05). There were no four-way interactions for positive words (ps > 0.12).
Discussion
The purpose of the study was to investigate whether bilingual readers exhibit L2 emotional
word processing effects, guided by predictions from an embodied approach to L2 word representation.
We found that bilinguals reading in their L2 showed word processing facilitation by embodied
knowledge only for some types of words, presumably because they capitalize on some but not all
sources of experiential information to ground L2 semantics. Specifically, bilinguals processed positive
words faster than neutral words, suggesting that they capitalize on emotionally positive experiences.
Moreover, bilinguals with high L2 proficiency processed negative and neutral words faster when they
were concrete compared to abstract, suggesting they also capitalize on sensorimotor experiences.
However, bilinguals processed negative words faster than neutral words only for the latest first-pass
measure, suggesting that they do not as readily capitalize on emotionally negative experiences.
Previous work on L1 embodiment indicates that a concreteness advantage, where observed, is
diagnostic of emotional neutrality because it does not occur for emotionally charged words (Sheikh &
Titone, 2013). Thus, the concreteness advantage for negative words, coupled with the absence of an
early emotional advantage for negative words suggests that only negative words were emotionally
disembodied in our bilingual participants. The finding that negative words alone are emotionally
disembodied is consistent with research showing that bilinguals prefer to use their L2 less often than
their L1 in emotional contexts (Dewaela, 2004; Pavlenko, 2005). An L1 preference for emotional
contexts would reduce co-occurrence between words and emotional experiences, leaving words
99
emotionally disembodied in the L2. We see evidence of this in our data given that increased L2
proficiency among our bilingual participants correlated with increased L2 use in formal contexts like
the work place but not informal social exchanges (detailed in supplementary material in Appendix B).
However, the same is not true of positive words, possibly because a positivity bias ensures that L2 use
co-occurs with emotionally positive experiences. We also found that the emotional advantage for
positive words was larger for high-frequency items compared to low-frequency items for first fixation
duration only, which was unexpected, suggesting that bilinguals more easily retrieve positive emotional
semantics for high-frequency words than low-frequency words at the earliest stage of processing,
though this needs to be confirmed in future work.
The selective disembodiment for negative words observed here is consistent with Conrad et al.
(2011), who found that bilinguals who were less proficient in their L2 than L1 showed emotional
effects only for positive and not negative words. Crucially, we extend their findings by showing that
this valence asymmetry occurs during natural reading, at the earliest stages of comprehension.
Bilinguals are able to eventually compute the negative valence of disembodied words, which emerged
as a relatively late emotional advantage for negative words. The late advantage is consistent with
Harris et al. (2003) who observed attenuated skin conductance responses for taboo words and
childhood reprimands even though bilinguals were ultimately able to identify their negative valence on
a rating task. Interestingly, bilinguals continue to process negative and neutral words similarly in terms
of how negative and neutral words are influenced by frequency, concreteness, and proficiency, even
after the emergence of the late emotional advantage. Thus, the late emotional advantage for negative
words and the emotional advantage for positive words do not differ solely in terms of time-course.
The embodiment approach also suggests a potential explanation for why studies vary in whether
reduced or eliminated emotional effects are observed during L2 language processing. When the L2 is
at least equal in proficiency to the L1, there is no L1 preference that locks the L2 out from emotional
contexts, as that preference is partly driven by greater L1 proficiency (Dewaele, 2004). Thus,
100
bilinguals can ground words in emotional experiences (Eilola et al., 2007; Sutton et al., 2007). In
contrast, when bilinguals are less proficient in their L2, the L2 gets locked out of those contexts and
bilinguals end up with emotionally disembodied words (Harris et al., 2003; Segalowitz et al., 2008).
Of course, bilinguals could acquire words in emotional contexts if forced into emotional contexts
irrespective of proficiency—the present results do not preclude that possibility, and indeed, suggest a
future test of the embodiment hypothesis.
The findings also clarify the role of L2 proficiency. Specifically, L2 proficiency predicted
concreteness advantages but not emotional advantages presumably because sensorimotor experiences
are not context-specific the way emotional experiences are (Matsumoto et al., 2005). Moreover,
concreteness advantages were limited to low-frequency words at high levels of L2 proficiency, similar
to L1 readers (Sheikh & Titone, 2013). Thus, the bilingual experiences that underlie word processing
facilitation by frequency and concreteness seem less dependent on context than the experiences that
underlie facilitation by emotion. Interestingly, facilitation by concreteness at high levels of L2
proficiency did not occur for first fixation duration or single fixation duration. Thus, the
representational changes correlated with L2 proficiency do not appear to be activated on the very first
fixation on target words.
The present findings also add to the small number of studies for emotional target words
embedded in sentences using eye-movement measures of reading, though these previous studies all
examined L1 processing. Some work on L1 processing using isolated words found that valence had a
monotonic effect on word recognition (i.e., negative words were slower than neutral words, which were
slower than positive words; Kuperman, Estes, Brysbaert, & Warriner, 2014). However, other studies
found that people process negative and positive words faster than neutral words (e.g., Vinson, Ponari,
& Vigliocco, 2014), which converges with recent eye-movement results for low-frequency words
(Scott, O'Donnell, & Sereno, 2012; Sheikh & Titone, 2013). There is also an earlier eye-movement
study by Hyönä and Häikiö (2005), but emotionally charged words in that study were only presented
101
parafoveally and never directly fixated by readers, as in our study. This methodological difference
makes it difficult to directly compare the results across the studies. In contrast, Scott et al. (2012) used
a natural sentence reading paradigm, and the present findings converge with their results for highfrequency words. Scott et al. found that for low-frequency words, people processed negative and
positive words (which did not differ) faster than neutral words. For high-frequency words, positive
words were faster than neutral words, but negative words did not differ from neutral words. This is
precisely the pattern that we observed here for bilinguals. One possible explanation that Scott et al.
(2012) posited to explain their findings for high-frequency items involves the selective attenuation of
emotional charge for negative words. Specifically, they proposed that negative emotionality may be
reduced by high frequency of exposure, which they compared to desensitization in psychotherapy.
Thus, although the specifics differ, our interpretation converges with their suggestion in terms of
negative words not being as emotional as positive words.
To conclude, our findings demonstrate that bilinguals have emotionally disembodied negative
words during L2 reading, and that these words are instead grounded in sensorimotor experiences like
neutral words. Our study also shows that L2 proficiency predicts concreteness advantages but not
emotional advantages during natural reading. Thus, sensorimotor experiences are more readily
available than emotionally negative experiences for grounding L2 words. Similarly, emotionally
positive experiences are more readily available for grounding L2 words than emotionally negative
experiences.
102
PREFACE FOR CHAPTER 4
The study presented in Chapter 3 examined word processing in bilinguals (native speakers of
French) that read English sentences in their L2. Of particular interest was how emotional,
sensorimotor, and linguistic contributions to representation jointly shape word processing (indexed by
emotional valence, concreteness, and word frequency, respectively). The research questions focused on
early and late stages of word processing, which were measured with early and late eye-movement
measures of reading. The study also compared the bilinguals with the native speakers of English from
Chapter 2 that read the same sentences in their L1. The study in Chapter 3 found that bilinguals read
positive words faster than neutral words across early and late eye-movement measures. However,
bilinguals read negative words faster than neutral words only during later stages of processing. In
addition, bilinguals read concrete words faster than abstract words among negative and neutral words,
but not positive words. In contrast, the native speakers of English in Chapter 2 read negative and
positive words faster than neutral words across early and late measures, and the concreteness advantage
was limited to neutral words. Thus, negative words are emotionally disembodied during L2 processing
given that they pattern with neutral words during early stages of processing. This emotional
disembodiment is consistent with research which shows that bilinguals lock their L2 out of contexts
that afford emotional experiences (Conrad, 2011; Dewaele, 2004; Pavlenko, 2005, 2012).
The study in chapter 4 follows up on a particularly salient processing difference between
negative and positive words across the studies in Chapters 2 and 3. Specifically, facilitation by positive
emotion was found for early measures in both studies. However, facilitation by negative emotion was
limited to late measures for bilinguals reading in their L2. A careful analysis of previous work on
emotional word processing during sentence reading suggests that the early effects of negative vs.
positive valence may differ when the target word varying on emotionality is preceded by text that is
more difficult to process (see Introduction in Chapter 4). This possibility also suggests that differences
between negative vs. positive valence may depend on whether people parafoveally process emotional
103
words before directly fixating them. Thus, a new set of sentences was generated for the study in
Chapter 4 to test the idea that early emotional facilitation might also be limited to positive words during
L1 processing, given the right circumstances. Moreover, the study in Chapter 4 also examined whether
this facilitation would vary as a function of word frequency and the opportunity to parafoveally process
target words before they are directly fixated using the boundary paradigm (Rayner, 1975).
104
CHAPTER 4:
The parafoveal and foveal effects of negative and positive emotional embodiment:
Evidence from eye movements and the gaze-contingent boundary paradigm
(Sheikh & Titone, 2015b, submitted for publication to Emotion)
105
Abstract
Prior work suggests that the embodied representation of emotional words causes them to be
differentially processed compared to neutral words during language processing, particularly when they
are abstract. However, an open question in this literature pertains to whether embodiment based on
negative or positive emotion play the same roles in language representation and processing. Here, we
tested whether negative and positive embodiment produce different effects during natural reading,
where at a given fixation people encode words that are currently fixated and to-be-fixated words in the
parafovea. Fifty native English speakers read emotional or neutral nouns in sentences. Target nouns
(angel) were preceded by definite articles, which were preceded by verbs (e.g., depicted the angel).
Parafoveal preview of nouns was valid or invalid before readers' eyes crossed a boundary between the
definite article and noun. The results showed that when people read the verb (depicted), fixation
durations for all first-pass measures were slower for valid nouns that were both emotional and low
frequent (when nouns were neutral, noun frequency did not influence reading times on the verb).
When people moved their eyes to the noun, however, gaze durations were faster when there had been a
valid preview of positive nouns that were high frequent. In contrast, an emotional advantage for
negative nouns was limited to a later stage of processing indexed by total reading time. Thus,
embodiment based on negative emotion took longer to facilitate foveal processing of the noun than
embodiment based on positive emotion.
106
Introduction
Recent theories of embodied word representation propose that words are semantically grounded
by the experiences they co-occur with during language use (Zwaan, 2008). However, an often cited
problem with embodied theories of cognition concerns how to ground intangible abstract concepts,
which cannot be directly experienced. For example, consider the words apple and grief. Apples are
concrete tangible objects that can be physically manipulated in any number of ways (e.g., sliced,
grasped, eaten, etc.). In contrast, grief is a less tangible and abstract concept that has indirect physical
manifestations (e.g., a lump in one’s throat). To address this question, it has been proposed that words
with concrete tangible referents (e.g., table) are grounded in sensorimotor experiences, and words with
abstract intangible referents (e.g., justice) are grounded in emotional experiences (Vigliocco, Meteyard,
Andrews, and Kousta, 2009). Thus, emotional experiences that are linked to words potentially
addresses the problem of how to ground intangible abstract concepts in the brain’s systems for
perception, action, and introspection (Barsalou, 1999). Consistent with this view, recent work suggests
that language processing is systematically affected when words are grounded in emotional experiences
vs. sensorimotor experiences, for both native speakers processing words in their first language (L1;
e.g., Kousta, Vigliocco, Vinson, Andrews, & Del Campo, 2011; Newcombe, Campbell, Siakaluk, &
Pexman., 2012; Sheikh & Titone, 2013) and bilinguals processing words in their second language (L2;
e.g., Sheikh & Titone, 2015; see also Ponari et al., 2015).
Of relevance here, while the idea of emotional embodiment provides some resolution to the
grounding problem for abstract semantics, several open questions remain. In this paper, we are
particularly interested in the role of negative vs. positive emotional embodiment (i.e., polarity
asymmetry). Of note, some studies find that people treat negative and positive words (e.g., shadow and
angel, respectively) the same way relative to neutral words (e.g., echo) (e.g., Sheikh & Titone et al.,
2013; Vinson, Ponari, and Vigliocco, 2014), whereas other studies find differences as a function of
valence polarity (i.e., negative vs. positive valence, e.g., Kuperman, Estes, Brysbaert, & Warriner,
107
2014; Scott, O’Donnell, & Sereno, 2012; Sheikh & Titone, 2015). Such discrepancies may reflect a
variety of factors, including differences in lexical attributes like word frequency and/or differences in
task demands. Thus, in the present study, we test whether a polarity asymmetry occurs when native
speakers read emotional target words in English sentences with no goal other than comprehension, with
the hope of determining whether negative and positive embodiment have different effects on L1
reading. In what follows, we first review studies based on words presented in isolation and then
describe eye-movement studies of sentence reading that motivated the current study.
One source of complexity regarding emotional word processing concerns the many other
linguistic ways that words vary (e.g., frequency), which may not have been adequately controlled in
past studies. For example, in a meta-analysis of 32 emotional Stroop studies, which typically show
slower responses for emotional vs. neutral words, Larsen, Mercer, and Balota (2006) found that
emotional words were longer in length and/or lower in frequency than neutral words (and/or had some
other lexical confound), and these confounds could account for different responses to emotional
compared to neutral words. Indeed, in a large-scale analysis of lexical decision and naming data from
the English Lexicon Project (ELP; Balota et a., 2007), Larsen et al. (2006) found that people did not
respond differently to emotional vs. neutral words after controlling for length, frequency, and other
lexical attributes. Thus, it is possible that past studies examining emotional word processing may have
found differences for emotional and neutral words because of reasons other than emotional valence.
While subsequent studies addressed the methodological concerns raised by Larsen et al. (2006),
the role of valence polarity (i.e., negative vs. positive emotional valence) remains unclear. For
example, using the Affective Norms for English Words (ANEW; Bradley & Lang, 1999), both Estes
and Adleman (2008) and Larsen et al. (2008) found different effects for negative and positive valence
in lexical decision and naming data from the ELP. Estes and Adleman (2008) found that responses
were generally slower for negative words compared to positive words. However, Larsen, Mercer,
Balota, and Strube (2008) tested interactions between valence and arousal, unlike Estes and Adleman,
108
and found that only negative words with particular arousal profiles elicited slower responses compared
to positive words. In contrast, other work tested participants in a lexical decision experiment and found
no differences between negative and positive valence (Kousta, Vinson, & Vigliocco, 2009). Rather,
people responded equally faster to negative and positive words compared to neutral words. Also of
note, the authors replicated this finding using ELP data and found that this valence effect did not
depend on arousal (Kousta et al., 2009). The findings in Kousta et al. (2009) were also replicated in
Vinson et al. (2014) using lexical decision data from the British Lexicon Project (Keuleers, Lacey,
Rastle, & Brysbaert, 2012).
Another source of complexity in the interpretation of past studies concerns possible interactions
between emotional valence and word frequency. For example, both Kuchinke, Võ, Hofmann, and
Jacobs (2007) and Scott, O'Donnell, Leuthold, and Sereno (2009) showed that lexical decisions for
low-frequency words were faster for negative and positive words compared to neutral words, and there
were no differences between negative and positive words. But for high-frequency words, only positive
words elicited faster lexical decisions than neutral words. Thus, Kuchinke et al. and Scott et al. (2009)
found a polarity asymmetry for high-frequency words. In contrast, Kuperman et al. (2014) found a
polarity asymmetry for low-frequency words. Kuperman et al. (2014), using ELP data for the largest
set of items to date, found that among low-frequency words, positive words elicited faster responses
compared to negative and neutral words, negative words elicited slower responses compared to neutral
words, and the emotional effects for negative and positive words vs. neutral words were attenuated for
higher frequency words.
A final issue that is unclear in past studies is the ecological validity of emotional valence
effects, that is, does emotional valence influence word processing when people read words as part of
sentences that are presented all at once. This issue was addressed by recent eye-movement studies of
reading where target words varying on emotionality were embedded in emotionally neutral sentences.
For example, Scott et al. (2012) had people read both negative and positive words faster than neutral
109
words when those words were low in frequency. But for high-frequency words, only positive words
were read faster than neutral words. Sheikh and Titone (2013) also found faster reading for negative
and positive words compared to neutral words among low-frequency items, and this emotional
advantage was maximal for words low in concreteness (i.e., abstract words) that are less likely to be
grounded in sensorimotor experiences. However, Sheikh and Titone (2013) found no emotional
differences for high-frequency items. In another recent study, Knickerbocker, Johnson, and Altarriba
(2015) found that people read negative and positive words faster than neutral words, similar to Sheikh
and Titone (2013) and Scott et al. (2012) for low-frequency words, though they did not test interactions
with frequency. Finally, in a recent study of Chinese sentence reading, Yan and Sommer (2015) found
that people read negative and positive words faster than neutral words, and that the emotional
advantage for negative words was weaker among higher frequency words.
Of relevance to the present study, Sheikh and Titone (2013) did not find different effects for
negative and positive valence in participants reading English sentences in their first language (L1
readers), however, they did find a polarity asymmetry in second language readers (L2 readers) that read
the same English sentences (Sheikh & Titone, 2015). The L2 readers were native speakers of French
and more dominant in their L1 than their L2, and read positive words faster than neutral words across
all first-pass reading time measures. However, the L2 readers read negative words faster than neutral
words only during the latest first-pass measure. In contrast, the L1 readers read negative and positive
words faster than neutral words across both relatively early and late first-pass measures.
Based on this pattern, one potential conclusion is that it is more difficult to activate embodied
semantics for negative words than for positive words when task demands are high (Sheikh & Titone,
2015), given that task demands would have been higher for the L2 readers that had less experience in
the stimulus language than the L1 readers. Thus, negative and positive valence might also have
different effects in L1 readers when sentence processing demands are increased. In partial support of
this idea, a comparison of the sentences in Sheikh and Titone (2013), who did not find a polarity
110
asymmetry, and the sentences in Scott et al. (2012), who did find a polarity asymmetry, suggests that
the identity of pretarget words might relate to differences between negative and positive valence.
Specifically, in Sheikh and Titone (2013), the target words varying on emotionality were preceded by
the pretarget words “about the” in approximately 65% of the sentences. In contrast, the most frequent
pretarget sequence in Scott et al. (2012), which was also “about the,” was repeated in approximately
7% of the sentences. The greater repetition of pre-target word sequences in Sheikh and Titone (2013)
may have globally reduced task demands for their participants at the point that they encountered the
target word, relative to what the participants experienced in Scott et al. (2012). Thus, sentences that
have unique pretarget words preceding the targets varying on emotionality may potentially increase
task demands and promote different effects for negative and positive valence.
Moreover, differences between negative and positive valence could depend on whether readers
process emotional target words before the targets varying on emotionality are directly fixated (i.e.,
while their eyes fixate preceding pretarget words), which can be tested using the gaze-contingent
boundary paradigm (Rayner, 1975). In the boundary paradigm, an invalid preview of a target word
(e.g., a nonword) is presented to readers while the target word region is in the parafovea (i.e., before it
has been fixated). A saccade crossing an invisible boundary (e.g., between the space before the target
word region and the preceding word) triggers a display change that replaces the invalid nonword with
the target word. Thus, while readers do not see the invalid nonword once the target word region is
brought into foveal vision with a direct fixation, they are presented with either a valid preview of the
target word or an invalid nonword while the target word region is in the parafovea. Previous work
shows that fixation durations on target words are approximately 30-50 ms shorter when a valid
parafoveal preview is available compared to when that preview is denied (i.e., a preview benefit; see
Rayner, 2009, for a review). Such preview benefits suggest that readers parafoveally process words
during reading (though the level of processing is subject to debate; Reichle, Pollatsek, & Rayner,
2006). Thus, if more cognitive resources are directed to emotional words because of their survival111
relevance, preview benefits on target words could be larger for emotional compared to neutral words.
It is also possible that reading times on pretarget words might differ for emotional vs. neutral
parafoveal words (i.e., emotion could modulate parafoveal-on-foveal effects). These emotional
modulations could also differ for negatively vs. positively valenced words.
The Present Study
To summarize, recent studies on emotional word processing generally find differences between
emotional vs. neutral words. The sentence reading studies, in particular, also make clear that emotional
valence influences word processing during natural reading, and is not an artefact of psycholinguistic
experimental manipulations. These emotional effects are consistent with the idea that embodied
emotional semantics become rapidly activated during word recognition and influence early stages of
word processing (Newcombe et al., 2012; see also Hino & Lupker, 1996; Pexman, Lupker, & Hino,
2002).
However, the studies reviewed above also produced variable findings, particularly with respect
to the role of emotional valence polarity. The reasons for different effects for negative vs. positive
valence are the focus of several theories. For example, some theories propose that responses are
prioritized for negative words compared to neutral and positive words because responses to negative
valence are more important to survival (Algom et al., 2004; Fox, Russo, Bowles, & Dutton, 2001;
Pratto & John, 1991). Others have theorized that emotional effects are more robust for positive words
compared to negative words (Conrad et al., 2011; Sheikh & Titone, 2015), potentially because a
positivity bias in human communication ensures that people experience more emotionally positive
experiences than negative experiences. Thus, these theories suggest that negative and positive
embodiment should, at least some of the time, produce different word processing effects.
In contrast, others argue that responses to negative and positive valence should not differ
(Kousta et al., 2009). This argument is based on the idea that behavioral avoidance and approach
systems that are presumably activated by negative and positive words, respectively, are equally
112
important to survival (Kousta et al., 2009). Thus, negative and positive embodiment should produce
the same word processing effects according to this perspective.
The work just reviewed leads to the following open questions regarding emotion effects when
people read in their L1: (1) Does emotional facilitation emerge earlier for positive words compared to
negative words, and do polarity differences vary as a function of word frequency? (2) Do people read
pretarget words differently as a function of parafoveal target words' emotional valence? (3) Do
preview benefits on target words vary as a function of their emotional valence? We address these
questions using eye-movement measures of sentence reading (Rayner, 2009), recorded when people
read semantically neutral sentences that contained English target nouns varying on emotional valence
and (continuously) on word frequency. Target nouns were preceded by the definite article, which was
preceded by a verb that was unique across sentences (e.g., the verb discerned and the noun angel in the
sentence, “The Mexican restaurant patrons discerned the angel that was represented in the painting in
the dining room.”). The definite article is typically skipped during reading (recent work suggests that
the definite article might even be automatically skipped when it is detected in the parafovea; Angele &
Rayner, 2013). Thus, we analyzed fixation durations on the verb and noun to test our hypotheses. We
also used the boundary paradigm to manipulate the parafoveal preview of the noun, which was
controlled by an invisible boundary in between the space preceding the noun region and the definite
article.
Native speakers of English read the sentences presented in their entirety with no goal other than
comprehension. In our previous study, we found polarity asymmetries for emotional facilitation at
early but not late measures (Sheikh & Titone, 2015). Thus, we analyzed all first-pass fixation time
measures from first fixation duration to go-past time to identify the points in time at which differences
emerge between the valence conditions.
113
Method
Participants
We tested 72 native speakers of English at McGill University that reported no history of
neurological or psychiatric disorder, and had normal or corrected-to-normal vision. We excluded
subjects that were high on alexithymia. Alexithymia is a personality construct which involves
difficulty processing emotion, and is a risk factor for a variety of psychiatric and psychosomatic
disorders (Bagby, Parker, & Tayler, 1994). We previously showed that individuals high on alexithymia
read emotional words differently compared to individuals low on alexithymia (Sheikh & Titone, 2013).
Thus, to insure that this factor would not play a role in the current study, 16 participants with
alexithymia scores >= 52.5 were excluded (Franz et al., 2008). We also excluded six participants that
had <= 2 observations in one or more experimental cell2. The analyses included the remaining 50
participants.
Materials
We used 288 target nouns that were selected from a database of words normed for emotional
valence and arousal (Kousta et al., 2009). The stimulus specifications are presented in Table 1. The
emotional categories differed on valence (negative < neutral < positive). For arousal, negative words
were higher than neutral words, and positive words were higher than negative words and neutral words.
We confirmed that the valence effects were not driven by arousal with models that included arousal as a
covariate. Frequency was manipulated but balanced across emotional categories, as was concreteness,
length, number of syllables, number of morphemes, and bigram frequency. The target nouns were
embedded in emotionally neutral sentences and were preceded by the definite article, which in turn was
preceded by a pre-target verb that was unique across sentences and varied on frequency (log HAL
frequency: M = 7.61, SD = 2.01) and length (M = 8.89, SD = 1.70). Verb frequency and length were
2
The results were similar if these six participants were included in the analysis. For the fixation duration measures for the
verb, some significant effects were reduced to trends. For the fixation duration measures for the noun, there was no
change in the pattern of significance.
114
statistically controlled in all models for the verb. Verb frequency was also statistically controlled in all
models for the noun. Note that six verbs (eyeballed, misremembered, referenced, sensationalized,
showcased, targeted) did not have log HAL frequencies in the ELP, to which we assigned a value of 0.
115
Table 1. Mean (Standard Deviations) For Lexical Attributes For Target Words And Their Latent Semantic Analysis (LSA) Values Within
Sentence Frames
Low frequency
High frequency
Negative
Neutral
Positive
Negative
Neutral
Positive
a
Valence
3.71 (0.89)
5.35 (0.25)
6.48 (0.54)
3.79 (1.02)
5.28 (0.23)
6.53 (0.68)
Arousala
4.76 (1.03)
4.44 (0.61)
5.16 (0.9)
5.13 (1.04)
4.28 (0.62)
5.38 (0.92)
b
Log frequency 7.78 (0.71)
7.78 (0.73)
7.84 (0.64)
9.43 (0.59)
9.42 (0.53)
9.44 (0.5)
Length
6.02 (1.73)
5.92 (1.92)
6.08 (1.72)
5.98 (1.73)
5.73 (1.77)
5.85 (1.74)
c
Concreteness
480.81 (75.38)
479.02 (72.75)
479.35 (77.38)
480.21 (67.62)
487.08 (59.95)
476.98 (70.83)
BGb
1662.69 (742.92)
1723.15 (918.47)
1622.04 (826.91)
1721.67 (850.22)
1632.33 (834.42)
1604.17 (893.66)
b
NMorph
1.29 (0.46)
1.31 (0.51)
1.33 (0.69)
1.27 (0.54)
1.25 (0.48)
1.31 (0.51)
NSyllb
1.77 (0.83)
1.85 (0.87)
1.83 (0.83)
1.71 (0.68)
1.75 (0.79)
1.73 (0.79)
LSA
0.002 (0.06)
0.011 (0.07)
0.011 (0.06)
0.007 (0.05)
-0.001 (0.05)
0.025 (0.07)
a
Kousta et al. (2009)
b
The English Lexicon Project (Balota et al., 2007)
c
MRC Psycholinguistic Database (Coltheart, 1981)
Note. BG = mean positional bigram frequency; NMorph = number of morphemes; Nsyll = number of syllables.
116
Note that emotional words are, by definition, emotionally incongruent with their emotionally
neutral carrier sentences, and thus unexpected relative to neutral words. This is important because
Holt, Lynn, and Kuperberg (2009) showed in an event-related potential (ERP) study that emotional
words elicited a larger N400 than neutral words embedded in the same emotionally neutral sentences.
They observed this enhanced N400 even after restricting analyses to items that were matched on
predictability and plausibility according to human ratings. Thus, to ensure that our results would not
reflect differential fit between target nouns and sentences, we made target nouns unpredictable within
sentences. As can be seen in Table 1, mean latent semantic analysis (LSA) values (for the negative,
neutral, and positive words in the sentential contexts) were low for the different conditions, but were
significantly higher for positive words compared to negative and neutral words for high-frequency
items (see Table 1). Thus, we confirmed that valence effects were not driven by this LSA difference
with models that included a covariate for the LSA values.
We also used an experimental design that allowed us to control for any potential differences in
fit by modeling the random effects for the target noun-sentence frame pairings using linear mixed
model (LMM) analyses (Bates, Maechler, & Dai, 2009). Specifically, the target nouns were grouped
into 96 triplets each consisting of a negative, neutral, and positive word. We created three sentence
frames for each triplet that were well-formed when combined with any of the triplet’s target nouns.
Target nouns and sentence frames were presented once in a given experimental list and the
combinations were counterbalanced across lists (see Table 2). We then used the properties of this
design to model the variability for the set of sentences in which the target nouns appeared. Thus, we
statistically controlled for any potential differences in fit across the valence categories, whether driven
by predictability, plausibility, or some unknown factor(s).
117
Table 2. Target Words Were Presented Within One Of Three Sentence Frames in a Given List, Which Were Rotated Around Their Triplet’s
Negative/Neutral/Positive Target Words Across Lists so That Targets Were Fully Crossed With Sentence Frames
Sentence frames and target nouns
The new employee researched the weapon/estate/novel that his boss read about and wanted to take a closer look at.
The elder scholar revealed the weapon/estate/novel that he had studied overseas in his latest research.
The university student examined the weapon/estate/novel that he would write about in his essay.
118
The parafoveal preview of the target noun was either valid or invalid. Display changes were
triggered once a saccade crossed an invisible boundary between the definite article and the space
preceding the noun (see Table 3). In the invalid condition, a nonword was presented in the parafovea
instead of the target noun, which matched the noun on word shape (using the approach described in
Angele, Slattery, Yang, Kliegl, & Rayner, 2008). Mean length from the space before the verb to the
invisible boundary before the noun was 8.89 (SD = 1.7) characters.
119
Table 3. Readers Were Provided With A Valid or Invalid Parafoveal Preview Of The Target Nouns Before A Saccade Crossed An Invisible
Boundary In Between The Definite Article And the Space Before The Noun, Marked Here By A Solid Line
Invalid parafoveal preview
Before boundary
The new employee researched the| mouqae that his boss read about and wanted to take a closer look at.
After boundary
The new employee researched the| weapon that his boss read about and wanted to take a closer look at.
Before boundary
After boundary
Valid parafoveal preview
The new employee researched the| weapon that his boss read about and wanted to take a closer look at.
The new employee researched the| weapon that his boss read about and wanted to take a closer look at.
120
Apparatus And Procedure
Eye movements were recorded using an SR Research Eyelink 1000 eye-tracker. The right eye
was recorded but viewing was binocular. Sentences were displayed on a 21-in. Viewsonic CRT
monitor with 144 Hz refresh rate, 1024 × 768 resolution, yellow 10-point Monaco font on black
background, with three characters subtending approximately 1° of visual angle. Yes or no questions
followed 21% of sentences to ensure participants read for comprehension. Questionnaires were
administered after the reading task, which asked participants about their language background (and
confirmed that they were native speakers of English) and also asked whether they noticed the display
change manipulation.
Results
We analyzed the following first-pass fixation duration measures of reading (Rayner, 2009):
First fixation duration (FFD; the duration of the first fixation on a word), single fixation duration (SFD;
the fixation duration in cases where the word was fixated exactly once), gaze duration (GD; the sum of
the durations of all fixations made during the first pass, before the eyes left the word region), and gopast time (GPT; the sum of the durations of all fixations on the word from the point when the word is
first fixated up until the eyes move past the word to the right).3 We also analyzed total reading time for
the noun to follow-up on a potentially interesting late effect that was hinted at in the analysis of the
first-pass measures. Fixations less than 80 ms were discarded unless they were within 1 character
space of adjacent fixations, with which they were combined.
The eye-movement measures were analyzed in R (R Development Core Team, 2010) with
linear mixed models (LMMs) using the lme4 package (Bates et al., 2009). In all models, we specified a
three-way interaction between valence (negative vs. neutral vs. positive), noun frequency (continuous),
and preview (correct vs. incorrect) in the fixed-effect structure. The baseline was set to correct preview
3
We also analyzed fixation probability (the probability of fixating a target on the first pass). There
were no significant emotional effects on fixation probability for the verb or the noun (ps > .14).
121
and neutral valence because our hypotheses specifically concern differences between negative and
positive valence vs. neutral valence. Noun frequency was centered by the mean. All models included
covariates for concreteness and the frequency of the verb. The models for the noun data also included a
covariate for noun length, and the models fit to the verb data included a covariate for verb length. All
covariates were continuous variables centered by their means.
We maximized the random-effect structure to the extent possible (Barr et al., 2013; see also
Bates et al., 2015). The models fit to the verb data included the following random effects structure.
For FFD, we specified random intercepts for participant (subject), word, and sentence; and by-subject
and by-sentence random slopes for preview. For SFD, we specified random intercepts for subject,
word, and sentence; and by-subject and by-word random slopes for preview. For GD, we specified
random intercepts for subject and sentence and by-subject and by-sentence random slopes for preview.
For GPT, we specified random intercepts for subject and sentence; and by-subject and by-sentence
random slopes for preview. The models fit to the noun data included the following random effects
structure. For FFD, we specified random intercepts for subject and word and by-subject and by-word
random slopes for preview. For SFD, we included random intercepts for subject, word, and sentence;
and by-subject, by-word, and by-sentence random slopes for preview. For GD we included random
intercepts for subject, word, and sentence; and by-subject and by-word random slopes for preview. For
GPT, we included random intercepts for subject, word, and sentence; and by-subject and by-word
random slopes for preview. Of note, it was not possible to specify random slopes for valence.
We used likelihood ratio tests to confirm that the highest-order significant effects in each model
were justified to ensure that the data were not overfitted. The models were fit to log-transformed
observations (the pattern of significance is similar for the untransformed data) because the logtransformed observations meet the assumption of normal distribution of residuals, and non-log
predicted values were calculated for plotting partial effects. We report the estimated coefficient (b),
standard error (SE), and t value. An absolute value of t ≥ 1.96 falls within the 95% confidence interval
122
and is significant. For convenience, we calculated p values using an upper bound for the degrees of
freedom (number of observations – the number of fixed effects in the model) which should not lead to
anti-conservative values (see Baayen, 2008). As can be seen below, the p values are consistent with the
|t| ≥ 1.96 criterion.
Trials with invalid display changes (e.g., completed during fixations rather than saccades,
triggered by drifts landing to the left of the boundary, etc.) were discarded (17.07%). Trials were
excluded if there was a track-loss (e.g., due to blinks) or critical regions were not fixated (20% for
nouns and verbs). We excluded outliers for the individual measures that were more than 2.5 standard
deviations from a subject’s means in the preview cells (FFD [1.90%], SFD [1.81%], GD [1.57%], and
GPT [1.44%] for the noun; FFD [2.09%], SFD [2.00%], GD [1.87%], and GPT [1.65%] for the verb).
Mean accuracy on the comprehension questions was 88.3% (SD = 4.5), indicating that participants
were engaged in the task of reading for comprehension. Means and standard deviations for FFD, SFD,
and GD, broken down by valence, noun frequency, verb frequency, and preview are presented in Table
4 for the verb and Table 5 for the noun. We first present an analysis of trials where the definite article
between the verb and noun was skipped (about 79% of the trials) since the definite article is typically
skipped during reading. We then test whether a fixated definite article interacts with preview.
123
Table 4. Means (standard deviations) for eye-movement measures for the verb
Valid preview
Low frequency
High frequency
Negative
Neutral
Positive Negative
Neutral
Positive
First fixation duration
264 (81) 270 (85) 261 (80) 264 (81) 270 (85) 261 (80)
Single fixation duration 275 (75) 281 (80) 270 (73) 275 (75) 281 (80) 270 (73)
Gaze duration
342 (146) 346 (146) 337 (147) 342 (146) 346 (146) 337 (147)
Go-past time
364 (162) 377 (168) 359 (163) 364 (162) 377 (168) 359 (163)
Invalid preview
Low frequency
High frequency
Negative
Neutral
Positive Negative
Neutral
Positive
First fixation duration
265 (79) 264 (83) 269 (83) 265 (79) 264 (83) 269 (83)
Single fixation duration 273 (76) 273 (80) 278 (82) 273 (76) 273 (80) 278 (82)
Gaze duration
345 (149) 338 (147) 343 (144) 345 (149) 338 (147) 343 (144)
Go-past time
367 (162) 363 (167) 369 (165) 367 (162) 363 (167) 369 (165)
Table 5. Means (standard deviations) for eye-movement measures for the noun
Valid preview
Low frequency
High frequency
Negative Neutral
Positive
Negative Neutral
Positive
First fixation duration
261 (69) 260 (67) 258 (67) 261 (69) 260 (67) 258 (67)
Single fixation duration 271 (68) 271 (68) 266 (64) 271 (68) 271 (68) 266 (64)
Gaze duration
306 (110) 309 (109) 290 (99) 306 (110) 309 (109) 290 (99)
Go-past time
333 (123) 336 (125) 317 (115) 333 (123) 336 (125) 317 (115)
Invalid preview
Low frequency
High frequency
Negative
Neutral
Positive Negative
Neutral
Positive
First fixation duration
270 (80) 269 (77) 264 (77) 270 (80) 269 (77) 264 (77)
Single fixation duration 286 (81) 282 (76) 281 (79) 286 (81) 282 (76) 281 (79)
Gaze duration
320 (121) 314 (114) 323 (127) 320 (121) 314 (114) 323 (127)
Go-past time
364 (147) 353 (137) 353 (137) 364 (147) 353 (137) 353 (137)
124
Early Reading Measures on the Verb
As can be seen in Table 6, there were three-way interactions between preview, negative valence,
and noun frequency, and preview, positive valence, and noun frequency for FFD, SFD, GD, and GPT.
There were no differences between negative and positive nouns in models with negative valence as the
baseline. The partial effects for SFD are plotted in Figure 1. Separate models fit to the SFD data split
by preview showed that there was a trend for a two-way interaction between negative valence and noun
frequency (b = -0.02, SE = 0.01, t = -1.77, p = .08), and a two-way interaction between positive valence
and noun frequency (b = -0.03, SE = 0.01, t = -2.41, p < .05) in the correct preview condition, but not in
the incorrect preview condition. These results, combined with the partial effects in Figure 1, suggest
that SFD on the verb was longer for low-frequency emotional nouns compared to high-frequency
emotional nouns when there was a valid preview of the noun, but that noun frequency did not influence
reading times on the verb when the parafoveal noun was emotionally neutral. For the other measures,
similar patterns were found in the partial effects and the models split by preview (though the effects in
the models fit to the GPT data split by preview generally seemed weaker, and trends were present in the
models for the correct preview data and the models for the incorrect preview data).
125
Table 6. Test of interaction between preview, valence, noun frequency for first pass fixation durations on the verb
First fixation duration Single fixation duration
Gaze duraiton
Go-past time
Fixed effect
b
SE
t
b
SE
t
b
SE
t
b
SE
t
(Intercept)
5.55 0.02 337.2*** 5.61 0.02 262.63*** 5.78 0.03 217.26*** 5.86 0.03 202***
Prev
-0.01 0.01 -1.34
-0.01 0.01 -0.87
-0.01 0.02 -0.46
-0.02 0.02 -1.21
Neg
0
0.01 -0.03
0
0.01 0.3
0
0.01 -0.24
-0.01 0.01 -1.01
Pos
0
0.01 -0.31
0.01 0.01 0.53
0
0.01 0.01
-0.01 0.01 -0.99
NFreq
0.01 0.01 1.2
0.01 0.01 1.23
0.01 0.01 1.26
0.01 0.01 0.73
Conc
0
0
0.11
0
0.01 0.32
0
0.01 -0.14
0
0.01 -0.38
VFreq
-0.02 0.01 -4.31*** -0.03 0.01 -5.46***
-0.04 0.01 -6.07***
-0.05 0.01 -6.71***
VLen
0
0.01 -0.12
0.01 0.01 1.86
0.04 0.01 6.55***
0.05 0.01 5.97***
Prev × Neg
0
0.01 0.11
0
0.02 -0.13
0.01 0.02 0.45
0.01 0.02 0.66
Prev × Pos
0.01 0.01 0.92
0
0.02 0.2
0
0.02 -0.1
0.02 0.02 0.81
Prev × NFreq
-0.01 0.01 -0.67
-0.02 0.01 -1.45
-0.01 0.01 -0.99
-0.01 0.01 -1.05
Neg × NFreq
-0.03 0.01 -2.68**
-0.02 0.01 -2.02*
-0.02 0.01 -1.45
-0.02 0.01 -1.56
Pos × NFreq
-0.03 0.01 -2.47*
-0.03 0.01 -2.47*
-0.03 0.01 -2.19*
-0.03 0.01 -1.93
Prev × Neg × NFreq 0.04 0.01 2.6**
0.03 0.02 1.99*
0.04 0.02 2.16*
0.05 0.02 2.56*
Prev × Pos × NFreq 0.03 0.02 1.96*
0.04 0.02 2.4*
0.05 0.02 2.39*
0.04 0.02 2.12*
Note. Prev = preview; Neg = negative; Pos = positive; NFreq = noun frequency; Conc = concreteness; VFreq = verb frequency;
VLen = verb length
* p < 0.050, ** p < 0.010, *** p < 0.001
126
Figure 1. Verb single fixation duration partial effects as a function of preview, valence, and noun
frequency after removing the effects of noun concreteness, verb frequency, verb length, and betweensubject, between-word, and between-sentence variance. Error bands show 95% confidence intervals.
127
Early Reading Measures on the Noun
As can be seen in Table 7, there was a main effect for preview for FFD, SFD, and GPT,
indicating that reading times on the noun were longer when there was an invalid preview of the noun
compared to when there was a valid preview (as noted in Table 7, the main effect of preview for FFD
was significant after dropping the nonsignificant interactions with valence and noun frequency). For
GD, there was a three-way interaction between preview, positive valence, and noun frequency. The
partial effects for GD are plotted in Figure 2. There were no differences between negative and positive
nouns in models with negative valence as the baseline.
128
Table 7. Test of interaction between preview, valence, noun frequency for first pass fixation durations on the noun
First fixation duration
Single fixation duration
Gaze duration
Go-past time
Fixed effect
B
SE
t
b
SE
t
b
SE
t
b
SE
t
(Intercept)
5.54 0.02 344.69***
5.58 0.02 314.22*** 5.69 0.02 280.31*** 5.78 0.02 261.53***
Prev
0.02 0.01 1.44†
0.03 0.01 2.32*
0.02 0.02 1.39
0.04 0.02 2.79**
Neg
0
0.01 -0.13
-0.01 0.01 -0.54
-0.02 0.01 -1.15
-0.02 0.01 -1.25
Pos
-0.02 0.01 -1.53
-0.01 0.01 -1.25
-0.03 0.01 -1.92
-0.02 0.01 -1.17
NFreq
0
0.01 -0.65
-0.01 0.01 -0.94
0
0.01 -0.23
0
0.01 -0.29
VFreq
0
0
-0.2
0
0
-1.25
-0.02 0
-3.45***
-0.03 0.01 -4.89***
Conc
0
0
0.75
0
0
1.09
0
0
0.11
0
0.01 0.47
Len
-0.01 0
-2.4*
0
0
1.17
0.04 0
9.62***
0.06 0.01 10.95***
Prev × Neg
0
0.01 -0.12
0
0.02 -0.04
0.02 0.02 0.92
0.03 0.02 1.38
Prev × Pos
0.01 0.01 0.5
0
0.02 0.28
0.04 0.02 1.93
0.03 0.02 1.31
Prev × NFreq
0
0.01 -0.14
0
0.01 0.09
-0.02 0.01 -1.3
0
0.01 -0.04
Neg × NFreq
-0.01 0.01 -0.79
0
0.01 0.18
-0.01 0.01 -0.99
-0.01 0.01 -0.52
Pos × NFreq
0
0.01 -0.42
-0.01 0.01 -0.8
-0.04 0.01 -2.59**
-0.04 0.01 -2.95**
Prev × Neg × NFreq 0.01 0.01 0.5
0
0.02 0.23
0.02 0.02 1.22
0
0.02 0.17
Prev × Pos × NFreq
0
0.01 -0.04
0.01 0.02 0.89
0.05 0.02 2.29*
0.03 0.02 1.56
Note. Prev = preview; Neg = negative; Pos = positive; NFreq = noun frequency; VFreq = verb frequency; Conc = concreteness;
Len = length; † The effect of preview for FFD was significant after dropping the nonsignificant interactions with valence and noun
frequency (b = 0.02, SE = 0.01, t = 2.43, p < .05)
* p < 0.050, ** p < 0.010, *** p < 0.001
129
Figure 2. Noun gaze duration partial effects as a function of preview, valence, and noun
frequency after removing the effects of noun concreteness, noun length, and verb frequency, and
between-subject, between-word, and between-sentence variance. Error bands show 95%
confidence intervals.
130
Separate models fit to the GD data split by preview showed that there was a two-way
interaction between positive valence and noun frequency (b = -0.04, SE = 0.01, t = -2.52, p < .05)
in the correct preview condition, but not in the incorrect preview condition. These results,
combined with the partial effects in Figure 2, show that GD on the noun was shorter for positive
nouns compared to neutral nouns, and that this emotional advantage became larger as noun
frequency increased.
For GPT, there was a two-way interaction between positive valence and noun frequency.
This interaction was also significant when positive valence was contrasted with negative valence
in a model with negative valence as the baseline (b = -0.03, SE = 0.01, t = -2.44, p < .05). Thus,
given that positive nouns differed from negative and neutral nouns in the full model, we fit
separate models to the GPT data split by median noun frequency with positive valence as the
baseline. In the model fit to the data for low-frequency nouns, GPT was shorter for negative
nouns compared to positive nouns (b = -0.04, SE = 0.02, t = -2.00, p < .05). In the model fit to
the data for high-frequency nouns, GPT was shorter for positive nouns compared to negative
nouns (b = 0.04, SE = 0.02, t = 1.99, p < .05) and neutral nouns (b = 0.05, SE = 0.02, t = 2.72, p
< .01). Thus, the advantage for positive words that was found for GD, particularly for higher
frequency positive words, occurred irrespective of preview during the relatively later stage of
processing indexed by GPT.
Total Reading Time On The Noun
Although there were no effects for negative valence for reading times on the noun, we
can see in Table 6 that the t statistic for the main effect of negative valence became larger across
the relatively earliest measure (FFD) to the latest measure (GPT). To follow up on this
potentially interesting pattern across the measures, we fit a model to the total reading time (TRT)
131
data for the noun, which is the sum of all fixation durations on the word, including fixations
made after the first pass. There was a three-way interaction between preview, positive valence,
and noun frequency (b = 0.05, SE = 0.02, t = 1.99, p < .05), which was similar to the three-way
interaction for GD on the noun that we discussed above. More interestingly, there was also a
main effect for negative valence showing that TRT was shorter for negative nouns compared to
neutral nouns (b = -0.05, SE = 0.02, t = -2.45, p < .05). There were no differences between
negative and positive nouns in a model with negative valence as the baseline. Thus, reading
eventually became faster for negative words compared to neutral words with sufficient
processing time.
Fixated Definite Article
In the previous sections, we analyzed the eye-movement data for the majority of trials
where the definite article between the verb and noun was skipped, which is typical of reading
and constituted the majority of trials in the present study (79%). Indeed, previous work shows
that the definite article is skipped even when it is grammatically/semantically incongruous with
the sentential context, suggesting that it is automatically skipped during reading (Angele &
Rayner, 2013). Thus, fixations on the definite article are potentially mislocated, and could have
been intended for another word (e.g., the subsequent noun; see Schotter, Angele, & Rayner,
2012). The results above confirm that emotional nouns parafoveally influenced processing,
while readers fixated the verb, on trials where the definite article was not fixated. The results
also confirm that preview benefits on the noun interacted with the emotional valence of the noun
on trials where the definite article was not fixated. Thus, these parafoveal and preview effects
were not limited to rare trials where the definite article received (potentially mislocated)
fixations.
132
However, the trials where the definite article was fixated, though few in number, are
potentially interesting with respect to reading times on the noun, since fixations on the definite
article would bring a reader maximally close to the noun. Thus, we tested whether simply
including the trials where the definite article was fixated changed the pattern of effects for the
noun. Including these trials did not change the pattern of significance. However, it is possible
that fixations on the definite article modulate the effect of the preview manipulation of the noun:
Since the boundary that triggered the display change was in between the definite article and the
space preceding the subsequent noun, fixations on the definite article would bring the reader
maximally close to a valid preview of the noun in the valid condition, or an invalid preview of
the noun in the invalid condition before a display change replaced it with the valid target. Thus,
we tested the interaction between a predictor that codes whether the definite article was fixated
on a given trial, which we will refer to as “the”-fixation (the baseline was set to no fixation on
the definite article), and preview for FFD, SFD, GD, and GPT. The analysis for FFD showed a
significant interaction between preview and the-fixation (b = 0.14, SE = 0.01, t = 10.3, p < .001).
We then fit separate models to the data split by preview, which showed that the benefit for a valid
preview was larger when the definite article was fixated (b = -0.08, SE = 0.01, t = -8.7, p < .001),
and that the cost of an invalid preview was larger when the definite article was fixated (b = 0.06,
SE = 0.01, t = 5.74, p < .001). The pattern of effects was the same across the other measures.
Thus, fixations on the definite article magnified the cost of an invalid preview and the benefit of
a valid preview.
Discussion
Guided by recent work on bilinguals reading sentences in their L2 (Sheikh & Titone,
2015), the purpose of the present study was to investigate whether native speakers of English
133
reading sentences in their L1 exhibit different word processing effects for embodiment based on
negative vs. positive valence,. We examined whether such differences emerge for target nouns
that varied on emotionality and frequency, and tested whether such differences arise parafoveally
while readers fixated a verb that preceded the noun or foveally once the noun was directly
fixated. We found that as readers fixated the verb, they showed emotional parafoveal-on-foveal
effects when they had a valid preview of the noun. Specifically, the verb was read faster as the
noun increased in frequency, but only when nouns were emotional, and there were no differences
between negative and positive nouns. However, readers showed foveal word processing
differences for negative vs. positive nouns only once they fixated the noun. Specifically, the
noun was read faster when it was emotionally positive compared to neutral during the first pass,
though this emotional advantage was maximal for high-frequency nouns and specific to the valid
preview condition. In contrast, negative nouns were read faster than neutral nouns only after the
first pass during a later stage of processing, irrespective of preview and noun frequency. Thus, in
the present set of sentences, embodiment based on negative emotion took longer to facilitate
foveal processing than embodiment based on positive emotion, which runs counter to our
previous work on L1 emotional word processing (Sheikh & Titone, 2013; see also Kousta et al.,
2009; Vinson et al., 2014).
In recounting the results, we first discuss the different foveal effects for negative vs.
positive valence, given that the small number of sentence reading studies that examined
emotional word processing focused on foveal processing. We then follow up with a discussion
of the emotional parafoveal-on-foveal effects, which were identical for negative and positive
valence.
134
Foveal Processing
Our previous work suggests that a late emotional advantage during foveal processing,
where observed, reflects difficulty in activating embodied semantics based on emotion (Sheikh &
Titone, 2015). Here, the emotional advantage for negative nouns was limited to a late stage of
processing reflected by total reading time on the noun. In contrast, the emotional advantage for
positive nouns was observed during an early stage of processing reflected by gaze duration on
the noun. Thus, readers more easily activated embodied semantics based on positive emotion
than negative emotion. A slower effect for negative emotion has also been observed in research
showing that bilinguals have difficulty activating embodiment based on negative emotion
(Sheikh & Titone, 2015; see also Conrad et al., 2011). In some respects, the slower effect for
negative emotion also parallels work showing that emotional advantages manifest more readily
for positive words than negative words across word frequency profiles (Kuchinke et al., 2007;
Scott et al., 2009, 2012; Yan & Sommer, 2015). Thus, one conclusion that can be drawn from
the previous and current findings is that emotional advantages are more robust for positive words
than negative words. However, it is not clear what mechanism underlies this effect.
One possibility is that emotional advantages for negative words get attenuated or
eliminated by regulatory mechanisms that suppress emotionally negative stimuli so that
emotional reactions to them do no disrupt ongoing behavior. Evidence of such mechanisms has
been found in work on emotion regulation. For example, Sheppes, Scheibe, Suri, and Gross
(2011) showed that people can block emotionally negative stimuli at an early stage of processing
after becoming aware of their negative emotion. The findings in Scott et al. (2009) suggest that
such mechanisms may also influence word processing. Scott et al. (2009) found that lexical
decisions were faster for positive words than neutral words irrespective of word frequency.
135
However, lexical decisions for negative words were faster than neutral words only among lowfrequency items. Interestingly, ERPs gathered during the lexical decisions showed that the P1
component (80-120 ms), which reflects early sensory stages of processing, was smaller in
amplitude for negative words compared to both neutral words and positive words among highfrequency items. Thus, the ERP data showed that these negative words were easier to process
than neutral words during early sensory stages of processing, even though there was ultimately
no emotional advantage in the behavioral data for these negative words. One potential
explanation considered by Scott et al. (2009) focused on the role of regulatory mechanisms like
those proposed in perceptual defense theory (McGinnies, 1949) and the mobilizationminimization hypothesis (Taylor, 1991). These theories propose that organisms can block or
dampen the impact of negative emotion, presumably so that strong emotional reactions do not
derail behavior. In Scott et al. (2009), this behavior consisted of lexical decisions.
Of relevance here is discussion in Kuperman et al. (2014) on the role of emotion in
lexical decisions. Kuperman et al. argued that while emotional words (negative or positive) may
be processed faster than neutral words during early sensory stages of processing, emotionally
negative word meanings may interfere with the subsequent process of making lexical decisions.
For example, consider a model with semantic and orthographic units that represent embodied
knowledge based on negative emotion and linguistic knowledge based on orthographic
information, respectively. Furthermore, assume that lexical decisions are generated on the basis
of activation in the orthographic units. In the initial moments of word perception, activation may
flow from orthographic units to semantic units. After this initial flow of activation, semantic
units may then draw processing resources away from associated orthographic units (e.g., via
inhibitory connections), which would interfere with lexical decisions for negative words (e.g.,
136
Kuperman et al., 2014). However, other findings suggest that activation in semantic units can
also feedback to orthographic units in some circumstances (e.g., as in Pexman et al., 2002), and
facilitate lexical decisions for negative words (e.g., low-frequency words in Scott et al., 2009;
see also Kousta et al., 2009; and Vinson et al., 2014). Clearly, the variability in findings suggests
that more work is necessary to elucidate how negative meanings modulate lexical decisions, and
the factors that influence the direction of such modulations.
However, the idea that negative words can get suppressed in a manner that prevents them
from interfering with ongoing behavior is potentially useful for understanding how negative
emotion influences word processing during sentence reading. Recall that the pretarget verb was
unique across sentence frames in the present study, unlike the pretarget sequences in previous
English sentence reading studies of L1 emotional word processing (Scott et al., 2012; Sheikh &
Titone, 2013). For example, the pretarget sequence in Sheikh and Titone (2013) was repeated in
65% of the sentences. Pretarget sequences were also repeated in Scott et al. (2012), though less
often than in Sheikh and Titone (2013). Moreover, these studies did not control the grammatical
class of the pretarget word matching the verb in the present study: Both Scott et al. (2012) and
Sheikh and Titone (2013) included function words, which are relatively easier to process during
reading than the content words (i.e., the verbs) employed here (e.g., content words are fixated
about 85% of the time during reading, but function words are fixated only about 35% of the time;
Rayner, 2009). These factors would have decreased task demands for the readers in previous
studies at the point that they fixated the word varying on emotionality, relative to the readers
here, because processing difficulty for a given word, even if fixated, spills over to subsequent
words (e.g., Pollatsek, Juhasz, Reichle, Machacek, & Rayner, 2008).
Thus, readers in the present study may have suppressed the meanings of negative words at
137
an early stage of processing after becoming aware of them (Scott et al., 2009; see also Sheppes et
al., 2011), in order to prevent their emotionally negative meanings from interfering with verb
processing. The time course of the emotional effects is consistent with this picture: While
readers fixated the verb (presumably their first opportunity to process the meaning of the
subsequent noun), negative and positive embodiment had the same effect. However, once
readers fixated the noun, facilitation by negative emotion was delayed relative to facilitation by
positive emotion. This is precisely the time course that would be expected if readers selectively
suppressed negative nouns after becoming aware of them. However, an alternative explanation
of the delayed emotional advantage for negative nouns could be that it is generally more difficult
to activate embodiment based on negative emotion than positive emotion. This explanation
perfectly accounts for the pattern of effects on the noun measures. Moreover, it does not appeal
to the notion of suppression, making it simpler. However, this alternative explanation cannot
account for identical negative and positive valence effects on the verb measures.
The increased task demands in the present study also seem to have influenced the
emotional advantage for positive nouns, relative to what has been found in previous work. For
example, here, we found that the emotional advantage for positive nouns during foveal
processing was maximal among high-frequency items and specific to the valid preview
condition. However, in our previous study, we found no emotional advantages among highfrequency items during L1 emotional word processing (Sheikh & Titone, 2013), and other
studies found that emotional advantages for positive words did not vary as a function of word
frequency (Scott et al., 2012; Yan and Sommer, 2015). Another difference with past work is that
the time course of the emotional advantage for positive nouns during foveal processing was
relatively later than in previous studies. Here, it emerged first in gaze duration, which is an early
138
measure of first pass processing. However, in previous work, it emerged in first fixation duration
and single fixation duration, which are relatively earlier measures of first pass processing
(Knickerbocker et al., 2015; Scott et al., 2012; Sheikh and Titone, 2013; Yan and Sommer, 2015).
Interestingly, the findings in the present study for the L1 processing of positive nouns
converge in some respects with our previous work on L2 emotional word processing (Sheikh &
Titone, 2015). In the L2 study, we found that the emotional advantage for positive nouns was
larger for high-frequency items than low-frequency items, for first fixation duration only.
However, during relatively later first pass measures of word processing, the emotional advantage
for positive nouns was not modulated by frequency. We concluded that it may have been easier
for the L2 readers in the study to initially activate the embodied semantics for high-frequency
words compared to low-frequency words, given that they were reading in their non-dominant and
less proficient language. As we indicated above in our account for negative nouns, the task
demands for the readers here were likely higher than in previous studies of L1 emotional word
processing at the point that the readers fixated the noun that varied on emotionality. Thus, the
increased task demands may have caused the readers here to behave in some respects like L2
readers, which would explain why the valence by frequency interaction for positive nouns was
similar to what we previously found for L2 readers. Moreover, the relatively later time course
that we found here for positive nouns might also be expected, assuming that increased task
demands would have slowed down all processing at the point that readers encountered the noun.
Furthermore, this reasoning might also explain why the valence by frequency interaction for
positive nouns was specific to the valid preview condition: The cost of an invalid preview might
have further increased task demands to the point that the demands overrode the emotional
advantage for positive nouns. Alternatively, it is also possible that the foveal emotional
139
advantage for positive nouns emerges only if there is an opportunity to parafoveally process the
noun, and that the emotional advantage increased with noun frequency because high-frequency
nouns were more easily processed in the parafovea than low-frequency nouns. Future work is
will have to disentangle these two alternatives.
The foveal processing differences found here for negative and positive emotion extend
previous work on the emotional embodiment of L1 word representations (e.g., Vigliocco et al.,
2009). Thus far, such theories have treated negative and positive valence as two sides of the
same coin—both are thought to ground word meanings and exert the same effects on word
processing. This seems to be true at least some of the time, as we and others have previously
found (e.g., Kousta et al., 2009; Vinson et al., 2014; Sheikh and Titone, 2013). However, we
previously found that L2 readers may not capitalize equally on negative and positive emotion for
grounding semantics (Sheikh & Titone, 2015; see also Conrad et al., 2011). Moreover, we found
here that negative and positive embodiment can also have different effects in L1 word
processing, as have others using a variety of paradigms (e.g., Kuchinke et al., 2007; Kuperman et
al., 2014; Scott et al., 2009; 2012). Thus, it may be time for theoretical work on L1 embodiment
to consider the role of emotional polarity.
Parafoveal Processing
We now shift the discussion to parafoveal-on-foveal effects. There are two previous
sentence reading studies with which we can compare the present findings. Hyönä and Häikiö
(2005) presented emotional words as false parafoveal previews in a sentence reading task using
the boundary paradigm, and tested whether there were parafoveal-on-foveal effects for these
false previews, of which there was little evidence. However, they only used emotionally
negative words and did not manipulate word frequency, leaving open the possibility that
140
parafoveal-on-foveal effects might vary as a function of negative vs. positive valence or the
frequency of parafoveal emotional words. Moreover, Hyönä and Häikiö used taboo words,
which tend to be very low in frequency. Finally, they only presented emotional taboo words as
false previews that could not be fixated, which could have influenced the pattern of effects in
some global way. More recently, the Chinese reading study mentioned earlier (Yan & Sommer,
2015) also tested the idea that emotion modulates parafoveal processing, though the boundary
paradigm was not used to manipulate the opportunity to parafoveally process emotional target
words. They found that reading times on pretarget words were longer when parafoveal words
were emotionally positive words compared to emotionally neutral. They also found that reading
times on pretarget words were longer for emotionally negative parafoveal words that were high
in frequency compared to low in frequency. Their parafoveal-on-foveal effects differ from the
pattern observed here. However, the implications of their results for English sentence reading
are potentially unclear because, as they argued, the mapping between orthography and semantics
is more direct in Chinese than in alphabetic languages like English, which may allow Chinese
readers to extract more semantic information from the parafovea compared to English readers.
Indeed, the amount and kind of information that readers extract from the parafovea is
hotly debated, which brings us to a tangential but potentially interesting issue: Connections with
research on parafoveal-on-foveal effects during sentence reading in general. Recent work
suggests that lexical and semantic information is parafoveally processed during Chinese reading,
and this has been shown in a number of studies (Tsai, Kliegl, & Yan, 2012; Yan, Pan, Bélanger,
& Shu, 2014; Yan, Richter, Shu, & Kliegl, 2009; Yan, Zhou, Shu, & Kliegl, 2012; Yang, Wang,
Tong, & Rayner, 2012). However, lexical parafoveal-on-foveal effects in alphabetic languages
like English are more controversial (Schotter et al., 2012). We did not design our study to
141
address this controversy. Rather, our goal was to examine whether embodiment based on
negative and positive valence had different effects during L1 sentence reading, and to identify
the time course of such differences. As we discussed above, negative and positive emotion had
identical parafoveal effects while readers fixated the verb, but different foveal effects emerged
once the noun was fixated. Thus, the different foveal effects for negative and positive valence
were more interesting with respect to our hypotheses.
However, the identical parafoveal effects for negative and positive nouns vs. neutral
nouns are also interesting, particularly with respect to the controversy about parafoveal-on-foveal
effects in alphabetic languages (which are often directly linked to debates about whether words
are lexically processed during reading one at a time in a serial fashion or in parallel; e.g.,
Reichle, Rayner & Pollatsek, 2003; and Engbert, Longtin, & Kliegl, 2002; respectively; see
Schotter et al., 2012, for a review). Here, we found that when there was a valid preview of the
noun in the parafovea, fixation durations on the verb were shorter as the noun increased in
frequency, but only when the noun was emotionally negative or positive.
This parafoveal-on-foveal frequency effect was previously found in corpus studies of
reading that examined whether fixation durations on words were correlated with the frequency of
subsequent words (e.g., Kennnedy & Pynte, 2005; Kliegl, Nuthmann, & Engbert, 2006;
Wotschack & Kliegl, 2013; note that the term successor effect may be more appropriate than
parafoveal-on-foveal effect for the findings in these corpus studies; see Angele et al., 2015, for
discussion). However, a common criticism of these corpus studies is that their parafoveal-onfoveal effects may reflect mislocated fixations (Drieghe, 2011; Schotter et al., 2012). A
mislocated fixation occurs when a given word in the parafovea is targeted for processing and
direct fixation, but some other word (e.g., a preceding word) ends up being fixated instead
142
because of oculomotor error. In this situation, the intended target remains in the parafovea, and
critics argue that this parafoveal word then influences the fixation duration on the preceding
word not because the two words are being processed in parallel, but because the fixation on the
preceding word was meant for the parafoveal word. Thus, follow-up studies used experimental
manipulations to discount the possibility that their effects were driven by mislocated fixations
(e.g., Risse & Kliegl, 2012). We did not include such a manipulation here, since our goal was to
study processing differences between negative and positive words, not whether words are
processed serially or in parallel during reading. Thus, we cannot exclude the possibility that the
parafoveal-on-foveal effects found here were driven by mislocated fixations.
However, multiple factors make it unlikely that the parafoveal-on-foveal effects found
here reflect mislocated fixations. Recall that the parafoveal-on-foveal effects on the verb were
significant for first fixation duration and single fixation duration. Thus, the effects were not
driven by observations that primarily consisted of first-pass refixations on the verb that could
have been meant for the noun (as could be the case for gaze duration, because gaze duration is
the sum of multiple fixations made during the first pass). In principle, it is also possible for first
fixation duration and single fixation duration to consist of mislocated fixations. However, recall
that the verb and noun were separated by the definite article in the present study. Thus, the only
circumstance in which a first or single fixation duration on the verb could have been a misolcated
fixation intended for the noun would be if both the verb and the definite article were meant to be
skipped. However, it is very rare for readers to skip two adjacent words during reading (Angele
& Rayner, 2013), particularly here given that the verb was generally long, and long words are
almost always fixated (Rayner, 1998).
Thus, one could argue that it is unlikely that the parafoveal-on-foveal effects on the verb
143
in the first and single fixation duration data were driven by mislocated fixations. Of course, this
raises the question of why we observed parafoveal-on-foveal frequency effects, like some
researchers (e.g., Kennnedy & Pynte, 2005; Kliegl et al., 2006; Wotschack & Kliegl, 2013),
while others have not shown support for such effects (e.g., Angele & Rayner, 2011; Angele et
al., 2008; Henderson & Ferreira, 1993; Inhoff et al., 2000; see Drieghe, 2011, for a summary).
One factor may be the emotionality of the noun, as emotional words are thought to draw
attention (e.g., Pratto & John, 1991). Another factor may be the definite article in between the
verb and noun. Some work suggests that a parafoveal definite article automatically triggers
skipping (Angele & Rayner, 2013), which might have automatically directed attention to the
parafoveal noun to the right of the definite article, while readers fixated the verb preceding the
definite article. Another possibility is that readers became aware that the verb and noun were
being systematically manipulated across the sentences, given that they were always unique.
Thus, readers may have anticipated a new previously unseen noun whenever they fixated a new
previously unseen verb, and strategically prioritized the processing of these two words. This
possibility also suggests that the parafoveal-on-foveal effects found here may not necessarily
reflect parallel lexical processing. For example, readers may have serially shifted their attention
to the noun in the parafovea whenever they fixated a new previously unseen verb faster than is
generally the case because a previously unseen verb would cause them to anticipate a previously
unseen noun. Finally, it is also possible that some combination of these factors increased
parafoveal processing of the noun. For example, attention capturing emotional nouns together
with skip triggering definite articles (or some other combination of the factors just mentioned)
could have increased parafoveal processing of the noun as well. It is also possible that such
mechanisms are influenced by individual differences in sensitivity to emotion (e.g., Long &
144
Titone, 2007; Sheikh & Titone, 2013). These alternatives will have to be tested in future work.
To conclude, our findings demonstrate that the frequency of a noun parafoveally
influences word processing while readers fixate a preceding verb, when the noun is emotionally
negative or positive. Our study also shows that readers process negative and positive nouns
differently only once they directly fixate the noun after having had a valid parafoveal preview of
it. Thus, embodiment based on negative and positive emotion had different effects only during
foveal processing, after being processed identically during parafoveal processing.
145
CHAPTER 5:
GENERAL DISCUSSION
146
GENERAL DISCUSSION
The goal of the present dissertation was to investigate whether the emotional meanings of
words influence eye-movement behavior during sentence reading, guided by predictions from an
embodied approach to word representation. Previous work suggests that people should process
emotional words (e.g., angel) faster than neutral words (e.g., echo). However, several lines of
research suggest that this emotional facilitation should also be modulated by concreteness and
word frequency, and that negative and positive emotion should occasionally produce different
word processing effects. These ideas were the focus of the three studies presented in this
dissertation. Another goal in these studies was to determine whether emotional word processing
effects and embodied theories generalize to natural language processing, because prior work on
these topics is based almost entirely on response-based tasks that required laboratory-specific
responses.
In the present dissertation, the research questions in each study were tested using
sentence reading tasks and eye-movement measures of reading that did not impose any goal on
participants other than comprehension. Thus, this methodology allowed participants to behave
as they would when they read any other text, which is ideal to address the issue of ecological
validity. This eye-movement sentence reading paradigm is also the method of choice to examine
the time course of emotional word processing effects during sentence reading, another key
question tested by the three studies in this dissertation, because eye-movement measures of
sentence reading can be used to investigate emotional effects on early vs. late stages of word
processing. This exquisite temporal resolution was also used to elucidate the point at which
word processing was modulated by individual differences in sensitivity to emotion and
individual differences in language proficiency in Chapters 2 and 3, respectively. Moreover,
147
Chapter 4 also capitalized on both the combined temporal and spatial resolution of eye
movement recordings to determine whether differences between negative and positive valence
emerge during parafoveal or foveal word processing. Thus, the present dissertation leveraged the
unique strengths of the eye-movement reading methodology to address several hotly debated
topics on embodiment in word representation, and to connect research on emotional embodiment
with psycholinguistic work on eye movements during reading.
The following section summarizes the aims, methodology, and results of each study in
this dissertation. The implications of the primary research findings in this dissertation are then
discussed in the following section. Future work suggested by the present dissertation is then
addressed in the final section.
Summary of Studies
Chapter 2:
Sheikh, N. A., & Titone, D. A. (2013). Sensorimotor and linguistic information attenuate
emotional word processing benefits: An eye-movement study. Emotion, 13, 1107–1121.
doi:10.1037/a0032417
The primary aim of the study presented in Chapter 2 was to determine how embodied
contributions to word representation influence L1 word processing. To address this question, the
study examined how emotional, sensorimotor, and linguistic contributions to word representation
(indexed by emotional valence, concreteness, and word frequency, respectively) jointly
influenced word processing while participants read sentences in their L1. The novel contribution
of this study is that it demonstrates that emotional word processing is modulated by concreteness
and word frequency during sentence reading, and that eye movements during sentence reading
are modulated by individual differences in sensitivity to emotional and sensorimotor information.
148
Participants in this study were native speakers of English that read English sentences. The
sentences were emotionally neutral but included a target word that varied on emotionality
(negative vs. neutral vs. positive). The target words also varied on concreteness (continuous) and
word frequency (continuous). In each sentence, the target words followed and preceded the
words ‘the’ and ‘that’, respectively. Thus, the sentence frames forced a noun interpretation for
each target word. The target words were organized into triplets, each consisting of a negative,
neutral, and positive target word, and each target word was presented in one of three emotionally
neutral sentence frames written for that triplet. The sentence frames were well-formed when
combined with any of the triplet members (e.g., The art teacher presented the smoke/hill/drink
that the students were going to paint) and target word-sentence frame combinations were
counterbalanced. A questionnaire was used to measure individual differences in alexithymia,
which provided a continuous measure of individual differences in sensitivity to emotional and
sensorimotor information. Specifically, people with high alexithymia ratings have difficulty
processing emotional information and are hypersensitive to sensorimotor information, compared
to people with lower alexithymia ratings.
The analysis focused on both early stages of word processing indexed by measures of
how long the eyes fixated target words during the first pass (first fixation duration and gaze
duration), and later stages of word processing indexed by a measure of how long the eyes fixated
target words during the second pass (second pass time). The first set of analyses focused on
effects across all participants (i.e., without modeling the effect of alexithymia). During early
stages of processing, participants showed an emotional advantage that was modulated by
concreteness and word frequency, and a concreteness advantage that was modulated by emotion
and word frequency. Specifically, participants read emotional words faster than neutral words
149
among low-frequency items that had low concreteness ratings (i.e., abstract words), and there
were no differences between negative and positive words. Participants also read concrete words
faster than abstract words among low-frequency items that were emotionally neutral. In contrast,
during later stages of processing, processing was only modulated by emotion. Specifically,
participants read emotional words faster than neutral words irrespective of concreteness and
frequency. Thus, the processing effects of emotional embodiment were modulated by
sensorimotor and linguistic information during early stages of processing, but occurred
irrespective of sensorimotor and linguistic contributions to representation during later stages of
processing.
These patterns were modulated by individual differences in alexithymia, which indexes
difficulty processing emotional information and hypersensitivity to sensorimotor information.
For example, during early stages of processing, the concreteness advantage for low-frequency
words was smaller among individuals with higher alexithymia scores (i.e., people that are
hypersensitive to sensorimotor information). Alexithymia also modulated the emotional
advantage during early stages of processing, but differentially for negative and positive words.
Specifically, the emotional advantage for positive words compared to neutral words was smaller
among individuals with higher alexithymia scores (i.e., people that have difficulty processing
emotional information). In contrast, alexithymia had no effect on the emotional advantage for
negative words compared to neutral words. Despite the differential alexithymia effect on
negative vs. positive words, there were no early reading time differences between negative and
positive words, suggesting that alexithymia had a relatively small effect on positive words during
early stages of processing. In contrast, negative and positive words were processed differently as
a function of alexithymia during later stages of processing. Specifically, high alexithymia
150
reduced the emotional advantage for positive words compared to neutral words, particularly
among low-frequency items. Moreover, late processing became slower for positive words
compared to negative words as alexithymia increased. Thus, individuals with high alexithymia
scores had difficulty processing embodiment based on positive emotion, particularly during later
stages of processing.
The study presented in Chapter 2 makes a number of novel contributions to research on
emotional word processing and embodiment in word representation. First, the naturalistic
reading paradigm provided by the sentence reading task and eye movement recordings is more
ecologically valid than the response-based tasks predominately used in previous studies. Thus,
this study suggests that emotional word processing effects and embodied theories generalize to
natural language processing. Second, the eye-movement sentence reading paradigm in this study
was used to elucidate the time course of embodiment effects, which generally cannot be
straightforwardly examined in response-based tasks. Thus, this study was able to show that
embodiment differentially modulates early vs. late stages of word processing. Third, this study
also clarified how emotional, sensorimotor, and linguistic contributions to word representation
jointly influence word processing, which has several implications for embodied approaches to
word representation, which I will further discuss below.
Chapter 3:
Sheikh, N. A., & Titone, D. (2015a). The embodiment of emotional words in a second language:
An eye-movement study. Cognition and Emotion.
The primary aim of the study presented in Chapter 3 was to determine whether bilinguals
show the same embodiment effects for emotional words in their L2 as native speakers do in their
L1, or whether bilinguals have emotionally disembodied words in their L2. To address this
151
question, the study examined how emotional, sensorimotor, and linguistic contributions to word
representation (indexed by emotional valence, concreteness, and word frequency, respectively)
jointly influenced word processing in bilinguals that read sentences in their L2. The bilinguals
were native speakers of French that were more dominant and proficient in their L1 than their L2
(English). The sentences in this study were the same English sentences used in the study
presented in Chapter 2. Thus, it was possible to directly compare the bilinguals in this study that
read sentences in their L2 (L2 readers) with the native speakers in Chapter 2 that read the same
sentences in their L1 (L1 readers). The novel contribution of this study is that it examines the
role of embodiment in L2 sentence processing, and examines how individual differences in L2
proficiency relate to different sources of embodiment. The experimental design was essentially
identical to the study in Chapter 2, given that the L2 readers read the same sentences. However,
unlike the study in Chapter 2, the study in Chapter 3 included a self-report measure to assess
continuous individual differences in L2 proficiency across participants. The proficiency
construct is commonly employed in the L2 literature as a measure of experience with the L2, and
a continuous measure of this construct was included in this study to assess the role of proficiency
in L2 embodiment.
The analysis focused on first pass fixation duration measures, ranging from the earliest to
the latest measure (first fixation duration, single fixation duration, gaze duration, and go-past
time). For the earlier measures (first fixation duration, single fixation duration, and gaze
duration), people showed an emotional advantage for positive words but not negative words.
Specifically, people read positive words faster than neutral words, but there were no differences
between negative words and neutral words. Moreover, the emotional advantage for positive
words in the first fixation duration data (the earliest measure of first pass processing) was larger
152
among higher frequency items. People also read positive words faster than negative words
across all measures. An emotional advantage for negative words was found only in the go-past
time data (the latest measure of first pass processing), indicating that people read negative words
faster than neutral words only after they had more time to process negative words. Thus,
emotional advantages were not limited to low-frequency words, as they were for native speakers
reading in their L1 (Chapter 2), suggesting that the processing effects of emotional embodiment
occur irrespective of linguistic contributions to representation. Moreover, the emotional
advantage was reduced for negative words compared to positive words, suggesting that only
negative words were disembodied in the L2.
Regarding the role of L2 proficiency, only the two relatively latest measures (gaze
duration and go-past time) showed an effect for L2 proficiency. Specifically, bilinguals with
higher levels of L2 proficiency read negative and neutral words faster as target words became
more concrete, which reduced the emotional advantage for positive words. This concreteness
effect was maximal among low-frequency items and did not differ for negative vs. neutral words,
which further suggests that negative words were emotionally disembodied for these L2 readers.
In contrast to the L2 proficiency effect for negative and neutral words, proficiency had no direct
effect on reading times for positive words or the emotional advantage for positive words among
abstract items. Thus, L2 proficiency was related to sensorimotor embodiment, but not emotional
embodiment.
The results for the L2 readers suggest that emotional disembodiment is specific to
negative words. This conclusion was confirmed in an analysis that directly compared the L2
readers in this study and the L1 readers from the study in Chapter 2. Specifically, the joint
effects of emotional, sensorimotor, and linguistic contributions to representation were modulated
153
by language group only for negative words and not positive words. Thus, the direct comparison
between L2 readers and L1 readers also suggests that only negative words were emotionally
disembodied for the L2 readers.
The study presented in Chapter 3 makes a number of novel contributions to research on
L2 emotional word processing and embodiment in word representation. First, as was the case for
the study presented in Chapter 2, the naturalistic reading paradigm provided by the sentence
reading task and eye movement recordings is more ecologically valid than the response-based
tasks predominantly used in previous studies of L2 emotional word processing. The
investigation of emotional effects and concreteness effects in bilinguals using a sentence reading
task in this study is a novel contribution to research on embodiment. Thus, this study suggests
that emotional word processing effects generalize to natural language processing by bilinguals
reading in their L2. Second, this study also showed that embodiment effects based on negative
vs. positive emotion have different time courses for bilinguals reading sentences in their L2.
Third, this study also clarified the role of emotional, sensorimotor, and linguistic contributions to
word representation in bilinguals, and suggests that negative words are emotionally disembodied
in the L2. These findings are also novel contributions to research on embodiment and have
several implications for embodied approaches to word representation, which I will further
discuss below.
Chapter 4:
Sheikh, N. A., & Titone, D. A. (2015b). The parafoveal and foveal effects of negative and
positive emotional embodiment: Evidence from eye movements and the gaze-contingent
boundary paradigm. Manuscript submitted to Emotion.
The primary aim of the study presented in Chapter 4 was to determine whether
154
embodiment based on negative and positive emotion differentially influences L1 word
processing. To address this question, the study examined how emotionally negative, neutral, and
positive words are parafoveally and foveally processed in English sentences using the gazecontingent boundary paradigm. The novel contribution of this study is that it examines how
emotional words in English sentences influence parafoveal and foveal word processing.
Participants in this study were native speakers of English that read English sentences. The
design of this study is similar to the studies presented Chapters 2 and 3. For example, as in those
studies, target words that varied on emotionality (negative vs. neutral vs. positive) were
presented in one of three emotionally neutral sentence frames, the target word-sentence frame
combinations were counterbalanced, and the target words followed and preceded the words the
and that, respectively, which forced a noun interpretation for the target words. However, unlike
the previous studies, the word the was always preceded by a pretarget verb. Fixation durations
on the verb were analyzed to measure parafoveal processing of the target word varying on
emotionality, before it was directly fixated. Thus, for the study in Chapter 4, I refer to fixation
durations on the verb and fixation durations on the noun to indicate whether the discussion
concerns fixation durations on the pretarget verb or fixation durations on the target noun that
varied on emotionality, respectively. Also, I refer to parafoveal-on-foveal effects when the
emotionality of the noun parafoveally affected fixation durations on the verb (before the noun
was directly fixated) and foveal effects when the emotionality of the noun affected fixation
durations on the noun itself. As in the studies presented in Chapters 2 and 3, the target noun
varied on word frequency (continuous). However, unlike the previous studies, concreteness was
controlled rather than manipulated in order to control for any processing effects driven by
sensorimotor contributions to word representation. This was done because the primary question
155
in this study concerned differences between embodiment based on negative vs. positive
emotional information, not the role of embodiment based on sensorimotor information. In
addition, the parafoveal preview of the noun was manipulated using the gaze-contingent
boundary paradigm so that participants either had a valid or invalid preview of the noun before
they directly fixated it. This preview manipulation was used to determine whether emotional
modulations of parafoveal and foveal word processing depend on the opportunity to parafoveally
process the noun.
The analysis focused on first pass fixation duration measures, ranging from the earliest to
the latest measure (first fixation duration, single fixation duration, gaze duration, and go-past
time). An analysis of a later fixation duration measure that includes fixations made after the first
pass (total reading time), was also performed to follow up on a pattern that was hinted at in the
analysis of the first pass measures. The following is a summary of the results of these analyses.
While participants fixated the verb and there was a valid preview of the noun, they showed
emotional parafoveal-on-foveal effects that were similar across the first pass measures.
Specifically, participants read the verb faster as the parafoveal noun increased in frequency, but
only when nouns were emotionally negative or positive. Moreover, this emotional parafovealon-foveal effect on the verb did not differ for negative vs. positive nouns. However, participants
showed foveal word processing differences for negative vs. positive nouns in the valid preview
condition once the noun was directly fixated. Specifically, the gaze duration data for the noun
showed that participants read the noun faster when it was emotionally positive compared to
neutral during the first pass after having had a valid preview of the noun, and this emotional
advantage became larger among higher frequency nouns. In contrast, only the analysis of total
reading time, which reflects a later stage of processing after the first pass, showed faster
156
processing for negative nouns compared to neutral nouns. Moreover, this late emotional
advantage for negative nouns occurred irrespective of preview and noun frequency. Thus, for the
sentences in this study, embodiment based on negative emotion took longer to facilitate foveal
processing of the noun than embodiment based on positive emotion.
The study presented in Chapter 4 makes a number of novel contributions to research on
emotional word processing and embodiment in word representation. For example, this study
demonstrates that the emotional valence of English words modulates parafoveal word processing
during sentence reading. Moreover, this study suggests that embodiment based on negative vs.
positive emotion occasionally produce different word processing effects in L1 sentence
processing. Previous work on L2 emotional word processing suggests that embodiment based on
negative vs. positive emotion may produce different word processing effects when bilinguals
read sentences in their L2, which was investigated in the study presented in Chapter 3. However,
embodied approaches to word representation do not address different embodiment effects for
negative vs. positive emotion. Thus, this study suggests that future theoretical work on the topic
must consider asymmetries between negative and positive emotion, which I will further discuss
below.
Emotional Embodiment In Native Speakers
The work in the present dissertation represents an important contribution to research on
embodiment in word representation (Barsalou, 1999; Glenberg, 1997; Lakoff & Johnson, 1980;
Zwaan, 2008), and has particular significance to research on the role of emotional embodiment in
word representation. The role of emotion is an important gap in research on embodiment in
word representation, which was recently addressed by Vigliocco et al. (2009). Specifically,
Vigliocco et al. (2009) proposed that embodied information based on emotional experiences and
157
sensorimotor experiences, together with linguistic information, jointly represent words, and that
abstract and concrete words are mostly represented by emotional and sensorimotor information,
respectively. Moreover, Vigliocco and colleagues showed that embodied emotional and
sensorimotor information facilitate processing for abstract and concrete words, respectively
(Kousta et al., 2011). As I reviewed in the General Introduction, such facilitation is consistent
with a feedback activation framework in which feedback from rich semantics facilitates word
processing (Newcombe et al., 2012; see also Hino & Lupker, 1996; Pexman et al., 2007; Pexman
et al., 2002; Siakaluk et al., 2008; Siakaluk et al., 2008; Yap et al., 2011; Zdrazilova & Pexman,
2013). The theory in Vigliocco et al. (2009) provides an elegant solution to the often-cited
problem of how to ground abstract semantics (Barsalou, 1999; Glenberg, 1997; Lakoff &
Johnson, 1980), and has had a large impact on the field. However, the theory in Vigliocco et al.
(2009) also leaves open several questions. For example, how do emotional contributions to
representation (indexed by emotional valence) influence processing for words that vary in terms
of their sensorimotor contributions to representation (indexed by concreteness), or vice versa.
Moreover, how do linguistic contributions to representation (indexed by word frequency)
modulate processing effects driven by emotional and sensorimotor embodiment?
The work in this dissertation shows that in native speakers of English reading sentences
in their L1, embodied contributions to representation, whether based on emotional or
sensorimotor embodiment, functionally modulate processing only when linguistic information is
insufficient for generating a response (i.e., when word frequency is low) (Chapter 2). However,
embodiment does not functionally modulate behavior when rich linguistic information is
sufficient for generating a response (i.e., when word frequency is high). This finding is
consistent with the linguistic and situated simulation theory which proposes that embodiment
158
should not functionally modulate behavior when responses can be generated using only linguistic
information (Barsalou et al., 2008; see also Juhasz et al., 2011). The linguistic and situated
simulation theory is based on the idea that embodied information generally takes longer to
activate compared to linguistic information because linguistic information is more similar to the
linguistic stimuli that activate word representations (see also Tulving & Thomson, 1973). The
slower time course for embodied information would generally limit embodied modulations of
processing to situations where linguistic information is insufficient for generating a response.
Thus, the present dissertation sheds light on how the embodied and linguistic contributions to
representation proposed in Vigliocco et al. (2009) jointly influence processing. Moreover, the
present dissertation also identifies how the processing effect of one embodied source of
information (e.g., concreteness) is modulated by a different embodied source of information
(e.g., emotionality), when linguistic information is insufficient for generating a response.
Specifically, processing for the most concrete words, which are rich in sensorimotor information,
is not facilitated further from rich emotional information (i.e., negative or positive emotionality).
Conversely, processing for words that are embodied by emotion (i.e., negative and positive
words) is not facilitated further by rich sensorimotor information. Thus, the present dissertation
shows that any given embodied aspect of representation, whether based on emotional or
sensorimotor experience (and perhaps other types experiences that have not yet received
attention), will maximally influence processing when other aspects of representation, whether
embodied or linguistic, are insufficient for generating a response.
Another major goal in the present dissertation was to determine whether embodied
theories generalize to eye movements during sentence reading, which this dissertation has
confirmed. Thus, the results can be integrated with computational models of eye movements
159
during reading (e.g., Engbert, Longtin, & Kliegl, 2002; Reichle, Rayner, & Pollatsek, 2003).
Recall that saccades move the eyes from one word to the next in reading. According to some
models of reading (e.g., Reichle et al., 2003), the oculomotor system occasionally begins to
program a saccade out of a word after an initial analysis of that word indicates that word
recognition is imminent. The word recognition system then continues to process the word while
the oculomotor system programs the saccade to the following word. It is not yet clear what type
of information is processed during the initial analysis that initiates saccade programming. For
example, semantic information, linguistic information, or both, might be used to initiate saccade
programming (Reichle, Pollatsek, & Rayner, 2006).
Embodied and linguistic information may influence saccade programming during reading
in the following way. First, it is likely that every linguistic stimulus in a sentence activates the
linguistic information and the embodied emotional and sensorimotor information in its
representation, consistent with research on how word stimuli activate memory representations
across several levels (e.g., Barsalou, 2008; Pexman et al., 2002). When a word is high in
frequency, emotional and sensorimotor information do not functionally modulate processing
because linguistic information is sufficient to initiate a saccade program to the next word, before
embodied information is fully activated (Barsalou et al., 2008), though that embodied
information is still retrieved while the oculomotor system programs the saccade. However,
embodied information functionally modulates processing for harder-to-recognize low-frequency
words because a saccade program cannot be initiated until embodied information helps recognize
the word. Moreover, emotional and sensorimotor information interact when embodied sources
of information become necessary to initiate a saccade program. Specifically, for negative and
positive low-frequency words, rich emotional information is sufficient to program a saccade.
160
Thus, sensorimotor information does not provide any additional processing advantage (i.e., as
words become more concrete). Conversely, rich sensorimotor information associated with
concrete words maximally affects saccade programming when processing is not already
facilitated by emotional information (i.e., neutral words). Note that this interaction between
linguistic and embodied information applies to early stages of processing, when the language
system attempts to determine whether word recognition is imminent before initiating a saccade
program to the next word. During later stages, positive and negative words are processed faster
compared to neutral words irrespective of frequency and concreteness. The differential effects
on early vs. late stages of processing most likely reflect the different kinds of cognitive processes
that are engaged during early vs. late processing. For example, later stages of processing after
the first pass are more likely to reflect the integration of word meaning into sentence-level
representations, rather than the activation of word representations, which typically happens at an
early stage of processing during the first pass (Rayner, 1998; 2009). Thus, the findings for the
later stages of processing suggest that it takes less time to integrate emotional meanings than
neutral meanings into a global sentence-level context.
Emotional Embodiment in L2 Speakers
Another important gap in research on embodiment that I identified in the General
Introduction is whether embodied ideas generalize to the situation of when bilinguals process
words in their L2, given that research on embodiment in word representation is based almost
exclusively on native speakers processing words in their L1. One specific question that has
previously been discussed in the literature is whether bilinguals have “disembodied” L2 words
(Pavlenko, 2012). This question is based on sociolinguistic research which shows that bilinguals
prefer to use their L2 less often than their L1 in emotionally charged social contexts (Dewaela,
161
2004; Pavlenko, 2005). In embodied approaches to language, words are grounded in whichever
type of experience they tend to co-occur with during language use (Zwaan, 2008). Thus,
emotional words may be left disembodied if they are experientially locked out of emotional
social contexts. The present dissertation took this question further and asked whether
disembodiment might be specific to negative words. To address this question, I capitalized on
the research on L1 embodiment that was discussed above (Chapter 2). Specifically, the findings
in Chapter 2 suggest that a concreteness advantage, where observed, is diagnostic of emotional
neutrality because it does not occur for words that are emotionally negative or positive. Thus,
the concreteness advantage for negative words found in this dissertation for bilinguals reading in
their L2, coupled with the absence of an early emotional advantage for negative words Chapter 3
suggests that only negative words are emotionally disembodied in the bilinguals' L2. The
finding that negative words are emotionally disembodied is consistent with sociolinguistic
research which shows that bilinguals prefer to use their L2 less often than their L1 in emotionally
charged social contexts (Dewaela, 2004; Pavlenko, 2005). A preference among bilinguals for
using the L1 in emotional social contexts would reduce co-occurrence between L2 words and
emotional experiences. Thus, this preference may leave negative words emotionally
disembodied in the L2. However, the findings also show that emotionally positive words are not
disembodied in the L2. Positive words may be spared disembodiment because a positivity bias
ensures that L2 use, like L1 use, co-occurs with emotionally positive experiences. Another
interesting finding was that the emotional advantage for positive words was larger among higher
frequency items for the earliest measure of first pass processing (first fixation duration), which
was unexpected. Thus, bilinguals may also retrieve positive emotional semantics more easily for
high-frequency words than low-frequency words at the earliest stage of processing—though this
162
will need to be confirmed in future research.
The selective emotional disembodiment for negative words is consistent with the findings
of Conrad et al. (2011), who found that bilinguals who were less proficient in their L2 than L1
showed emotional effects only for positive and not negative words in ERP data recorded during
lexical decisions. Crucially, the findings in the present dissertation extend the work in Conrad et
al. by showing that this asymmetry between negative and positive valence occurs during
sentence reading when bilinguals process words with no goal other than comprehension, at the
earliest stages of word processing. Interestingly, the present dissertation also shows that
bilinguals are able to eventually compute the emotionally negative meanings of disembodied
words, which emerged as a relatively late emotional advantage for negative words compared to
neutral words. This late advantage is consistent with the findings in Harris et al. (2003), who
observed attenuated skin conductance responses in bilinguals for emotionally charged childhood
reprimands and taboo words, even though these bilinguals were ultimately able to identify the
negative valence of the reprimands and taboo words in an explicit rating task. Interestingly, the
present dissertation also found that even after the emergence of the late emotional advantage,
bilinguals continue to process negative and neutral words similarly in terms of how negative and
neutral words are influenced by frequency, concreteness, and proficiency. Thus, this later
emotional advantage for negative words and the emotional advantage for positive words that was
observed across early and later measures may not differ solely in terms of time-course.
An embodiment approach to word representation also provides an explanation for why
studies might vary in whether they find reduced or eliminated emotional effects in bilinguals
processing words in their L2. Specifically, in bilinguals that have an L2 that is at least equal in
proficiency to their L1, there should be no L1 preference that locks the L2 out of emotional
163
social contexts, because previous work shows that the L1 preference is partly driven by greater
proficiency in the L1 compared to the L2 (Dewaele, 2004). Such bilinguals should be able to
ground their L2 words in emotional experiences, given that their L2 is not locked out of
emotional contexts. Thus, they should not show evidence of emotional disembodiment (e.g.,
Eilola et al., 2007; Sutton et al., 2007). In contrast, when bilinguals are less proficient in their L2
than their L1, they will prefer to use their L1 for emotional contexts. Thus, such bilinguals will
lock their L2 out of emotional contexts and end up with emotionally disembodied L2 words (e.g.,
Harris et al., 2003; Segalowitz et al., 2008).
The findings also clarify the role of L2 proficiency in the embodiment of L2 words. For
example, the present dissertation found that L2 proficiency predicted concreteness advantages
but not emotional advantages. This difference presumably reflects that sensorimotor experiences
are less context-specific than emotional experiences (Matsumoto et al., 2005). Moreover, the
present dissertation also found that concreteness advantages were limited to low-frequency
words in bilinguals with higher levels of L2 proficiency. The modulation of the concreteness
advantage by word frequency is similar to what was found in this dissertation for native speakers
processing words in their L1 (Chapter 2). Thus, the bilingual experiences that underlie word
processing facilitation by word frequency and concreteness seem to be less dependent on social
context than the experiences that underlie word processing facilitation by emotionality.
Interestingly, facilitation by concreteness in bilinguals with higher levels of L2 proficiency did
not occur for the very earliest measures of first pass processing (first fixation duration and single
fixation duration). Thus, the changes in word representation that are correlated with L2
proficiency do not appear to be activated on the very first fixation on target words during L2
reading.
164
Embodiment Based on Negative vs. Positive Emotion
Another important gap in research on emotional embodiment is potential differences
between negative vs. positive emotion. Thus far, theories of emotional embodiment treat
negative and positive emotion as two sides of the same coin—both are thought to ground
semantic meanings and exert the same word processing effects (e.g., Vigliocco et al., 2009;
Kousta et al., 2011). However, as I reviewed in the General Introduction, the processing effects
of negative and positive valence are subject to intense debate: Some studies find that people
treat negative and positive words the same way relative to neutral words (e.g., Kousta et al.,
2009; Vinson et al., 2014), whereas other studies find differences as a function of valence
polarity (i.e., negative vs. positive valence, e.g., Kuchinke et al., 2007; Kuperman et al., 2014;
Scott et al., 2009, 2012).
The studies in the present dissertation also show that negative and positive valence
occasionally produced different word processing effects. For example, the study in Chapter 2
found that native speakers of English reading sentences in their L1 process negative and positive
words the same way relative to neutral words (though individual differences in processing
emotional information, as reflected by alexithymia scores, differentially modulate the processing
effects of negative vs. positive emotion). In contrast, the study in Chapter 3 found that bilinguals
that read the same sentences in their L2 showed different processing effects for negative and
positive words vs. neutral words. The results in Chapters 2 and 3 were based on measures of
foveal word processing (i.e., processing time during direct fixations on the target words varying
on emotionality). Thus, Chapters 2 and 3 suggest that emotional effects on foveal word
processing are modulated by whether the participants are L1 readers or L2 readers. Interestingly,
the study in Chapter 4 also found that people process negative and positive words differently
165
relative to neutral words, as in Chapter 3, even though it tested native speakers of English
reading sentences in their L1, like Chapter 2. Moreover, Chapter 4 used a different set of
sentences and methodology to show that polarity differences between negative and positive
emotion vary across measures of foveal word processing vs. parafoveal word processing (i.e.,
processing time during fixations in preceding regions of text, before the target words varying on
emotionality receive direct fixations). Specifically, Chapter 4 showed that people process
negative and positive words the same way relative to neutral words during parafoveal word
processing, but differently during foveal word processing. Thus, despite the fact that L1
processing was tested in Chapters 2 and 4, the foveal processing results in Chapter 4 were
actually more similar to the results for the L2 readers in Chapter 3. For example, in Chapter 4,
the foveal emotional advantage for negative words emerged later in processing than the
emotional advantage for positive words, as in Chapter 3. In some respects, the slower effect for
emotionally negative words in Chapter 4 also parallels work by other researchers who found that
emotional advantages manifest more readily for positive words than negative words across low
and high word frequency profiles during L1 word processing (Kuchinke et al., 2007; Scott et al.,
2009, 2012; Yan & Sommer, 2015).
Thus, results in Chapters 3 and 4 and studies by other researchers (e.g., Kuchinke et al.,
2007; Scott et al., 2009, 2012; Yan & Sommer, 2015) suggest that emotional advantages are
more robust for positive words than negative words. However, it is not clear what mechanism
underlies this effect. One possibility is that regulatory mechanisms might suppress emotionally
negative stimuli in order to prevent the disruption of ongoing behavior by emotional reactions to
the negative stimuli. This suppression would likely attenuate or reduce emotional advantages for
negative words. Research on emotion regulation has found evidence of such mechanisms. For
166
example, Sheppes, Scheibe, Suri, and Gross (2011) showed that people can block emotionally
negative stimuli at early stages of processing once they become aware of their negative emotion.
Scott et al. (2009) also found evidence for what appeared to be suppressed emotional effects for
emotionally negative words in their lexical decision results. In their discussion, Scott et al.
(2012) considered whether regulatory mechanisms like those proposed in perceptual defence
theory (McGinnies, 1949) and the mobilization-minimization hypothesis (Taylor, 1991) could
have suppressed negative words in their experiment, which would have eliminated emotional
effects for negative valence. The perceptual defence theory and the mobilization-minimization
hypothesis propose that people block negative stimuli so that strong emotional reactions to them
do not derail ongoing behavior. In Scott et al. (2009), that behavior consisted of lexical
decisions.
Of relevance here is recent discussion on the role of emotion in lexical decisions
(Kuperman et al., 2014). Kuperman et al. argued that while early sensory processing may be
faster for emotional words (negative or positive) than for neutral words, negative word meanings
may disrupt the subsequent process of making lexical decisions. For example, consider a model
in which embodied knowledge based on negative emotion and linguistic knowledge based on
orthographic stimuli are represented by semantic and orthographic units, respectively.
Furthermore, assume that the generation of lexical decisions is based on activation in the
orthographic units. In the initial moments of perception, the flow of activation might cascade
from orthographic units to semantic units. After the initial cascade of activation, semantic units
may then draw processing resources away from orthographic units (e.g., by reducing activation
in the orthographic units via inhibitory connections), which would disrupt lexical decisions for
negative words (e.g., Kuperman et al., 2014). However, other findings suggest that activation in
167
semantic units can also feedback to orthographic units in some circumstances (e.g., as in Pexman
et al., 2002), and facilitate lexical decisions for negative words (e.g., low-frequency words in
Kuchinke et al., 2007; and Scott et al., 2009; see also Kousta et al., 2009; and Vinson et al.,
2014).
It may help to consider how embodied approaches to word representation might integrate
with other models of word processing that focus on how performance is influenced by task
demands. Such work would not only reconcile embodied approaches with different models of
word processing, but could potentially elucidate different interactions between processes and
representations (Zwaan, 2014). Recent work has made progress on exactly this front. For
example, the feedback activation framework (Hino & Lupker, 1996; Pexman et al., 2002)
explicitly describes how semantic knowledge and decision criteria jointly influence task
performance. As I reviewed in the General Introduction, recent work has described how
embodiment influences word processing within this feedback activation framework, and this
work has been able to explain why embodiment based on sensorimotor vs. emotional experiences
produces differential effects on concreteness vs. abstractness decisions (Newcombe et al., 2012).
However, this approach presently also has limits as it has not been able to readily explain the
differential effects of embodiment based on negative vs. positive emotion on word processing
(Zdrazilova & Pexman, 2013). Clearly, more work is necessary to elucidate how negative
meanings modulate responses, given the variability in findings.
However, the idea that negative emotion can occasionally disrupt ongoing behavior (e.g.,
Kuperman et al., 2014), and the idea that regulatory mechanisms may be able to suppress
negative words to prevent such disruption (e.g., Scott et al., 2009), can potentially help explain
why embodiment based on negative vs. positive emotion occasionally produce different word
168
processing effects during sentence reading. Specifically, a comparison of the sentences used in
Chapter 4 vs. Chapters 2 and 3 suggests that the potential for disruption by negative emotion
may have been greater in the sentences in Chapter 4, because the pretarget sequences were more
difficult in Chapter 4 than in Chapters 2 and 3. This elevated difficulty potentially increased the
need for regulation. Recall that in Chapter 4, the pretarget verb was unique across sentence
frames, unlike the pretarget sequences in Chapters 2 and 3. Specifically, the pretarget sequence
in Chapters 2 and 3 included the same words in 65% of the sentences. Moreover, the
grammatical class of the pretarget word matching the verb in Chapter 4 was not controlled in
Chapters 2 and 3, which included a mix of content and function words. Function words are
relatively easier to process during reading than content words (e.g., content words and function
words are fixated about 85% and 35% of the time during reading, respectively; Rayner, 2009).
Thus, these factors may have increased task demands for the native speakers in Chapter 4 at the
point that they fixated the word varying on emotionality, relative to the native speakers in
Chapter 2, because processing difficulty for a given word spills over to subsequent words (e.g.,
Pollatsek et al., 2008). These factors may also explain why the results in Chapter 4 differed from
the L1 processing results in Scott et al. (2012) because the sentences in Scott et al. (2012)
differed from the sentences in Chapter 4 in the same ways discussed above.
Thus, participants in Chapter 4 may have suppressed the meanings of parafoveal negative
nouns after becoming aware of them (as in Scott et al., 2009; and Sheppes et al., 2011), so that
verb processing would not be disrupted by the emotionally negative meanings of the parafoveal
nouns. The time course of the emotional effects in Chapter 4 is consistent with this picture:
While participants fixated the verb (which was presumably their first opportunity to process the
subsequent noun), negative and positive embodiment had the same parafoveal word processing
169
effects. However, after participants directly fixated the noun, emotional facilitation of foveal
processing for negative nouns was delayed relative to emotional facilitation for positive nouns.
This is precisely the time course that one would expect if participants suppressed negative nouns
only after becoming aware of them, while the negative nouns were in the parafovea. An
alternative explanation of the delayed effects for negative nouns could be that embodiment based
on negative emotion is generally more difficult to activate compared to embodiment based on
positive emotion. This alternative perfectly explains the pattern of effects in the foveal word
processing results for the noun measures. Moreover, this alternative does not invoke the notion
of suppression. However, this alternative cannot account for the pattern of effects in the
parafoveal word processing results for the verb measures, which were identical for negative and
positive emotion.
The increased task demands in Chapter 4 also appear to have influenced the foveal
emotional advantage for positive words, relative to what was found in Chapter 2. Specifically,
the foveal emotional advantage for positive nouns in Chapter 4 was specific to the valid preview
condition and maximal among high-frequency items. However, there were no emotional
advantages among high-frequency items in Chapter 2, and other researchers found no word
frequency modulations of foveal emotional advantages for positive emotion (Scott et al., 2012;
Yan and Sommer, 2015). In addition, the time course of the foveal emotional advantage for
positive emotion was later than in Chapter 2. In Chapter 4, it emerged first in gaze duration,
which is an early measure of first pass processing. However, in Chapter 2, it emerged first in
first fixation duration, which is a relatively earlier measure of first pass processing than gaze
duration (see also Knickerbocker et al., 2015; Scott et al., 2012; Yan and Sommer, 2015).
Interestingly, the foveal word processing results for positive nouns in Chapter 4 also converge in
170
some respects with the L2 processing results in Chapter 3. In Chapter 3, the emotional
advantage for positive nouns was larger among higher frequency items in the first fixation
duration data, and I concluded that it might have been easier for the bilinguals to initially activate
the embodied semantics for high-frequency words compared to low-frequency words, given that
the bilinguals were reading in their non-dominant and less proficient language. This explanation
might also be true for the native speakers in Chapter 4. Specifically, the increased task demands
faced by the participants in Chapter 4 may have made it relatively more difficult to activate word
representations, as in an L2, even though they were reading in their L1, which would explain
why the valence by frequency interaction for positive nouns was similar to what was found for
L2 processing in Chapter 3. Moreover, the relatively later time course for emotional
modulations of foveal processing by positive valence in Chapter 4 might also reflect increased
task demands, assuming that increased demands would have slowed down all processing at the
point that participants encountered the noun. Furthermore, this reasoning potentially explains
why the valence by frequency interaction for positive nouns was specific to the valid preview
condition: The already high task demands may have been elevated further by the cost of an
invalid preview, to the point that the demands overrode the emotional advantage for positive
nouns. Alternatively, it is also possible that the foveal emotional advantage for positive nouns
emerges only if there is an opportunity to parafoveally process the noun, and that the emotional
advantage increased with noun frequency because high-frequency nouns were more easily
processed in the parafovea than low-frequency nouns. Future work is will have to disentangle
these alternatives.
Thus, the present dissertation shows that negative and positive emotion do not always
modulate processing the same way, as may be implied in theoretical work on emotional
171
embodiment (Vigliocco et al., 2009). Specifically, the possibility that negative and positive
emotion may be differentially involved in representation has not been addressed. Their roles in
representation and processing may be identical at least some of the time, as Chapter 2 suggests
(see also Knickerbocker et al., 2015; Kousta et al., 2009; Vinson et al., 2014). However, in other
instances, negative and positive emotion clearly produce different word processing effects, as
was found in Chapters 3 and 4 (see also Kuchinke et al., 2007; Kuperman et al., 2014; Scott et
al., 2009, 2012). Moreover, the processing effects of negative and positive emotion also differ
depending on whether representations are being parafoveally or foveally activated (Chapter 4).
Thus, it may be time for theoretical work on L1 embodiment to consider the factors that
modulate the roles played by negative and positive emotion in word representation.
Directions for Future Research
This dissertation provides the foundation and motivation for future investigations of
emotional embodiment in word representation. For example, this dissertation demonstrates that
eye movement recordings can be used to study emotional embodiment (and sensorimotor
embodiment) during natural language processing. Thus, eye movement recordings and sentence
reading tasks can be used to test questions about embodiment in an ecologically valid way, and
can also help test questions that cannot be easily addressed in response-based tasks. For
example, eye-movement reading paradigms can be used to test which representational features
are activated parafoveally vs. foveally during sentence reading (Rayner, 2009), as was done in
Chapter 4. However, in Chapter 4, reading times on a pretarget verb were always used to
measure parafoveal activation of a subsequent target word. Thus, an important question for
future work is whether the properties of the pretarget word modulate the representational features
that become parafoveally activated for the subsequent target word.
172
Eye-movement reading paradigms can also help study emotional word processing in
more elaborate contexts than sentences. Sentence reading tasks have gone a long way in
addressing concerns about ecological validity in response-based tasks. However, similar
concerns could potentially apply to sentence reading tasks given that sentence processing is
generally also embedded within a context. For example, consider the potential differences
between reading words in a sentence vs. a paragraph. The processing load in paragraph reading
is very likely more demanding than in sentence reading. The greater processing demands in
paragraph reading could potentially alter how the activation of word representations unfolds
relative to sentence reading. In support of this conjecture, previous work suggests that the
pattern of word frequency effects might vary as a function of the differential task demands in
sentence reading (Gollan et al., 2011) vs. paragraph reading (Whitford & Titone, 2012).
Using texts more extended than sentences as a vehicle to study emotional word
processing would also allow researchers to investigate questions that cannot be easily tested in
sentence reading paradigms. For example, a paragraph reading task could be used to test
whether emotional words exert more influence than neutral words on gist during skimming, and
whether emotional words capture more attention than neutral words during scanning. Such
issues and topics could be addressed in future research using eye-movement studies of paragraph
reading.
Another question for future research concerns the processing dynamics of emotionally
charged concrete words. A very small number of studies have examined how emotion and
concreteness modulate each other's processing effects in response-based tasks (Ferré, Ventura,
Comesana, Fraga, 2015; Kanske & Kotz, 2007; Newcombe et al., 2012). With regard to natural
reading, to the best of my knowledge, the studies in Chapters 2 and 3 are the only studies that
173
have examined how emotion and concreteness jointly modulate word processing. Recall that
Chapter 2 showed that words which benefit from an emotional advantage are not facilitated
further by increased concreteness (or conversely, words that benefit from a concreteness
advantage are not facilitated further by increased emotionality). Thus, another important
question to be addressed in future research is which representational content, emotional or
sensorimotor, underlies processing benefits for the particular subset of words that are both
emotionally charged and concrete. For example, if participants must strategically inhibit the
emotional semantics of concrete words in order to contend with the particular demands of a task,
would there still be a concreteness advantage for these words relative to abstract words? This
question could potentially be addressed in a reading study that manipulates the difficulty of
pretarget words preceding targets that vary on emotionality.
The work reported in this dissertation also suggests that future work should examine the
role of emotion regulation in emotional word processing. For example, a comparison of
Chapters 2 and 4 suggests that task demands modulate the foveal word processing effects of
negative vs. positive emotional embodiment in native speakers of English. However, this
conclusion was based on a comparison of stimulus characteristics across studies. Thus, an
additional topic for future research is whether task demands also modulate valence effects within
subjects, which could be addressed by manipulating the difficulty of the pretarget word
preceding the target that varies on emotionality. In addition, a study that examines the role of
individual differences in emotion regulation during word processing might also help identity the
circumstances in which people suppress the representations of emotionally negative words. For
example, individuals that generally regulate their emotional experiences by suppressing their
emotions, rather than engaging with them, may also be more likely to suppress emotionally
174
negative words in the context of an experimental task.
Another question for future work concerns the role of arousal in semantic representation.
In the present dissertation, emotion was operationalized in terms of a valence-based approach,
consistent with non-discrete approaches to emotion (Russell, 2003; Russell & Barrett, 1999), and
the focus was on embodiment based on negative and positive emotion. However, negative and
positive emotion can also vary in terms of intensity, as reflected by emotional arousal ratings. In
the present dissertation, additional analyses showed that valence effects do not depend on arousal
(see also Kousta et al., 2009; Kuperman et al., 2014; Vinson et al., 2014). However, one
methodological problem is that valence and arousal are not independent: Emotionally negative
and positive words tend to be more arousing than neutral words. Thus, testing their separate
effects is not trivial. Some studies have tried to factorially manipulate the valence and arousal of
words using relatively small stimulus sets and using repetition to obtain the necessary number of
trials, and these studies have found that arousing negative and positive words capture more
attention than less arousing negative and positive words (e.g., Anderson, 2005). Thus, it should
be possible to use a comparable approach to separate the effects of valence and arousal on word
processing during sentence reading.
A related question for future work concerns the structure of emotion. As I just mentioned
above, the present dissertation focused on negative and positive emotion, based on a non-discrete
approach to emotion (Russell, 2003; Russell & Barrett, 1999). In contrast, some researchers
have modeled emotions in discrete ways. For example, Ekman (2002) focuses on happiness,
sadness, fear, anger, and disgust. The theoretical work that motivated this dissertation took a
non-discrete approach to emotion that focused on valence (Vigliocco et al., 2009; Kousta et al.,
2011), as have most studies of emotional word processing (e.g., Eilola et al., 2007; Kousta et al.,
175
2009; Kuchinke et al., 2007; Kuperman et al., 2014; Scott et al., 2009, 2012; Sutton et al., 2007).
However, it has been argued that discrete emotions do not reduce down to brain patterns that can
be captured by variation on valence and arousal (e.g., Vytal & Hamann, 2010). Thus, it may be
worth considering whether basic emotions also play a role in word representation.
Another open topic concerns the motivational and experiential factors that underlie
emotional disembodiment in L2 speakers. For example, sociolinguistic work suggests that
emotional words are disembodied in an L2 because bilinguals prefer to use their L1 for
emotional social exchanges (e.g., Dewaele, 2004; Pavlenko, 2005). However, the specific
motivational factors that underlie these linguistic preferences receive little attention. Moreover,
it should be possible to elucidate the experiential factors that underlie disembodiment. For
example, is disembodiment related to the sense of familiarity bilinguals have for individual L2
words? In addition, are some emotional contexts more important for emotional embodiment than
others? Moreover, given that the sociolinguistic work implies that bilinguals choose to lock their
L2 out of emotional contexts, would negative words in the L2 remain disembodied if bilinguals
were forced into emotional contexts irrespective of proficiency? As I reviewed in the General
Introduction, embodied theories of word representation have virtually focused exclusively on
native speakers processing words in their L1. Thus, studies that tackle these questions would
represent significant contributions to the literature.
Another question for future research concerns the neural bases of embodiment in word
representation. As I argued in the General Introduction, it is unlikely that any given region of the
brain corresponds one-to-one with a word’s meaning. Instead, word meaning representations are
likely distributed across different regions of the brain and overlap with other words’ meanings.
Some researchers have argued that the amygdala may be part of a neural system that becomes
176
activated during emotional processing, particularly in tasks where there is an opportunity for
attentional mechanisms to be influenced by emotion (e.g., Vuilleumier & Huang, 2009).
Moreover, research using lexical tasks shows that the rostral anterior cingulate cortex is also
activated during emotional word processing (Vigliocco et al., 2013), which may modulate
amygdala activity during emotional word processing (see Etkin, Egner, Peraza, Kandel, Hirsch,
2006). The distributed networks that underlie the embodiment of word meanings have not yet
been fully realized (Kousta et al., 2011), and thus this topic is also a promising direction for
future investigations.
Conclusion
The research presented in this dissertation makes significant contributions to embodied
approaches to cognition and word representation. Ultimately, the findings suggest that linguistic,
sensorimotor, and emotional contributions to representation jointly modulate word processing
during natural reading, and that embodiment based on negative and positive emotion play
different roles in representation and word processing depending on the linguistic biographical
history of participants as well as online comprehension demands. Eye movement measures of
natural reading were used across all three studies, which provided a highly naturalistic and
ecologically valid test bed for embodied theories, which is a significant improvement upon the
response-based tasks that have been employed in previous research. Moreover, advanced linear
mixed model statistical techniques were used to test questions that are not easily addressed using
the standard statistical analyses that are typically used in embodiment research. Thus, the work
in this dissertation lays the groundwork for future research on how human beings represent word
meaning.
177
REFERENCES
Algom, D., Chajut, E., & Lev, S. (2004). A rational look at the emotional Stroop phenomenon: a
generic slowdown, not a Stroop effect. Journal of Experimental Psychology: General,
133, 323–338. doi:10.1037/0096-3445.133.3.323
Allen, R., & Hulme, C. (2006). Speech and language processing mechanisms in verbal serial
recall. Journal of Memory and Language, 55, 64–88. doi:10.1016/j.jml.2006.02.002
Altarriba, J. (2008). Expressions of emotion as mediated by context. Bilingualism: Language and
Cognition, 11, 165–167. doi:10.1017/S1366728908003295
Altarriba, J., Bauer, L. M., & Benvenuto, C. (1999). Concreteness, context availability, and
imageability ratings and word associations for abstract, concrete, and emotion words.
Behavior Research Methods, Instruments, & Computers, 31, 578–602.
doi:10.3758/BF03200738
Anderson, A. K. (2005). Affective influences on the attentional dynamics supporting awareness.
Journal of Experimental Psychology: General, 134, 258–281. doi:10.1037/00963445.134.2.258
Anderson, A. K., & Phelps, E. A. (2001). Lesions of the human amygdala impair enhanced
perception of emotionally salient events. Nature, 411, 305–309. doi:10.1038/35077083
Angele, B., & Rayner, K. (2011). Parafoveal processing of word n + 2 during reading: Do the
preceding words matter? Journal of Experimental Psychology: Human Perception and
Performance, 37, 1210–1220.
Angele, B., & Rayner, K. (2013). Processing the in the parafovea: Are articles skipped
automatically? Journal of Experimental Psychology: Learning, Memory, and Cognition,
39, 649–662. doi:10.1037/a0029294
178
Angele, B., Schotter, E. R., Slattery, T. J., Tenenbaum, T. L., Bicknell, K., & Rayner, K. (2015).
Do successor effects in reading reflect lexical parafoveal processing? Evidence from
corpus-based and experimental eye movement data. Journal of Memory and Language,
79–80, 76–96. doi:10.1016/j.jml.2014.11.003
Angele, B., Slattery, T. J., Yang, J., Kliegl, R., & Rayner, K. (2008). Parafoveal processing in
reading: Manipulating n + 1 and n + 2 previews simultaneously. Visual Cognition, 16,
697–707. doi:10.1080/13506280802009704
Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics using R.
Cambridge U. K: Cambridge University Press.
Bagby, R. M., Parker, J. D. A., & Taylor, G. J. (1994). The twenty-item Toronto Alexithymia
scale—I. Item selection and cross-validation of the factor structure. Journal of
Psychosomatic Research, 38, 23–32. doi:10.1016/0022-3999(94)90005-1
Bagby, R. M., Taylor, G. J., & Parker, J. D. A. (1994). The twenty-item Toronto Alexithymia
scale—II. Convergent, discriminant, and concurrent validity. Journal of Psychosomatic
Research, 38, 33–40. doi:10.1016/0022-3999(94)90006-X
Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical access?
The role of word frequency in the neglected decision stage. Journal of Experimental
Psychology. Human Perception and Performance, 10, 340–357.
Balota, D. A., Cortese, M. J., Sergent-Marshall, S. D., Spieler, D. H., & Yap, M. (2004). Visual
word recognition of single-syllable words. Journal of Experimental Psychology: General,
133, 283–316. doi:10.1037/0096-3445.133.2.283
Balota, D. A., Yap, M. J., Cortese, M. J., Hutchison, K. A., Kessler, B., Loftis, B., … Treiman, R.
(2007). The English lexicon project. Behavior Research Methods, 39, 445–459.
179
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for
confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68,
255–278.
Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577–
660.
Barsalou, L. W. (2008). Grounded Cognition. Annual Review of Psychology, 59, 617–645.
doi:10.1146/annurev.psych.59.103006.093639
Barsalou, L. W., Santos, A., Simmons, W. K., & Wilson, C. D. (2008). Language and simulation
in conceptual processing. In M. de Vega, A. M. Glenberg, & A. C. Graesser (Eds.),
Symbols, embodiment, and meaning (pp. 245–293). Oxford: Oxford University Press.
Barsalou, L. W., Simmons, W. K., Barbey, A. K., & Wilson, C. D. (2003). Grounding conceptual
knowledge in modality-specific systems. Trends in Cognitive Sciences, 7, 84–91.
doi:10.1016/S1364-6613(02)00029-3
Barsalou, L. W., & Wiemer-Hastings, K. (2005). Situating abstract concepts. In D. Pecher & R.
A. Zwaan (Eds.), Grounding cognition: The role of perception and action in memory,
language, and thinking (pp. 129–163). Cambridge, England: Cambridge University Press.
Bates, D., Kliegl, R., Vasishth, S., & Baayen, H. (2015). Parsimonious mixed models. Retrieved
from http://arxiv.org/abs/1506.04967
Bates, D., Maechler, M., & Dai, B. (2009). Lme 4: Linear mixed-effects models using S4 classes.
R package version 0.999375-27. Retrieved from http://lme4.r-forge.r-project.org
Bayer, M., Sommer, W., & Schacht, A. (2010). Reading emotional words within sentences: the
impact of arousal and valence on event-related potentials. International Journal of
Psychophysiology: Official Journal of the International Organization of
180
Psychophysiology, 78, 299–307. doi:10.1016/j.ijpsycho.2010.09.004
Benjafield, J. G., Kingstone, A., & Smilek, D. (2010). Cognition. Oxford University Press.
Binder, J. R., Westbury, C. F., McKiernan, K. A., Possing, E. T., & Medler, D. A. (2005). Distinct
Brain Systems for Processing Concrete and Abstract Concepts. Journal of Cognitive
Neuroscience, 17, 905–917. doi:10.1162/0898929054021102
Bleasdale, F. A. (1987). Concreteness-dependent associative priming: Separate lexical
organization for concrete and abstract words. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 13, 582–594. doi:10.1037/0278-7393.13.4.582
Boring, E. G. (1950). A History of Experimental Psychology. New York: Appleton-CenturyCrofts.
Boucher, J., & Osgood, C. E. (1969). The pollyanna hypothesis. Journal of Verbal Learning and
Verbal Behavior, 8, 1–8. doi:10.1016/S0022-5371(69)80002-2
Bradley, M. M., & Lang, P. J. (1999). Affective norms for English words (ANEW): Insruction
manual and affective ratings (Technical Report C-1). Gainesville: University of Florida,
Center for Research in Psychophysiology.
Brysbaert, M., & New, B. (2009). Moving beyond Kucera and Francis: A critical evaluation of
current word frequency norms and the introduction of a new and improved word
frequency measure for American English. Behavior Research Methods, 41, 977–990.
doi:10.3758/BRM.41.4.977
Burgess, C., & Livesay, K. (1998). The effect of corpus size in predicting reaction time in a basic
word recognition task: Moving on from Kučera and Francis. Behavior Research Methods,
Instruments, & Computers, 30, 272–277. doi:10.3758/BF03200655
Carpenter, S. K. (2005). Some neglected contributions of Wilhelm Wundt to the psychology of
181
memory. Psychological Reports, 97, 63–73. doi:10.2466/pr0.97.1.63-73
Chomsky, N. (1959). A review of B. F. Skinner’s “Verbal behavior.” Language, 35, 26-58.
Chomsky, N. (1968). Language and mind. New York: Harcourt Brace Jovanovich.
Chomsky, N. (1980). Rules and representations. Behavioral and Brain Sciences, 3, 1-61.
Chomsky, N. (1968). Knowledge and language. Berlin: Praeger.
Coltheart, M. (1981). The MRC psycholinguistic database. The Quarterly Journal of
Experimental Psychology Section A, 33, 497–505. doi:10.1080/14640748108400805
Conrad, M., Recio, G., & Jacobs, A. M. (2011). The time course of emotion effects in first and
second language processing: A cross cultural ERP study with German–Spanish bilinguals.
Frontiers in Psychology: Language Sciences, 2, 351. doi:10.3389/fpsyg.2011.00351
Cortese, M. J., & Schock, J. (2013). Imageability and age of acquisition effects in disyllabic
word recognition. Quarterly Journal of Experimental Psychology, 66, 946–972.
doi:10.1080/17470218.2012.722660
Decety, J., & Grèzes, J. (2006). The power of simulation: imagining one’s own and other’s
behavior. Brain research, 1079, 4–14. doi:10.1016/j.brainres.2005.12.115
Degner, J., Doycheva, C., & Wentura, D. (2012). It matters how much you talk: On the
automaticity of affective connotations of first and second language words. Bilingualism:
Language and Cognition, 15, 181–189. doi:10.1017/S1366728911000095
De Groot, A. M. (1989). Representational aspects of word imageability and word frequency as
assessed through word association. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 15, 824–845. doi:10.1037/0278-7393.15.5.824
Dewaele, J.-M. (2004). Blistering barnacles! What language do multilinguals swear in? Estudios
de Sociolinguistica, 83–105.
182
Dolan, R. J. (2002). Emotion, Cognition, and Behavior. Science, 298, 1191–1194.
doi:10.1126/science.1076358
Drieghe, D. (2011). Parafoveal-on-foveal effects in eye movements during reading. In S. P.
Liversedge, I. D. Gilchrist, & S. Everling (Eds.), Oxford handbook on eye movements (pp.
839–855). Oxford: Oxford University Press.
Duyck, W., Vanderelst, D., Desmet, T., & Hartsuiker, R. J. (2008). The frequency effect in
second-language visual word recognition. Psychonomic Bulletin & Review, 15, 850–855.
Eilola, T. M., Havelka, J., & Sharma, D. (2007). Emotional activation in the first and second
language. Cognition & Emotion, 21, 1064–1076. doi:10.1080/02699930601054109
Ekman, P. (1999). Basic Emotions. In T. D. R. Scientist & M. J. P. of C. Psychology (Eds.),
Handbook of Cognition and Emotion (pp. 45–60). John Wiley & Sons, Ltd.
Engbert, R., Longtin, A., & Kliegl, R. (2002). A dynamical model of saccade generation in
reading based on spatially distributed lexical processing. Vision Research, 42, 621–636.
doi:10.1016/S0042-6989(01)00301-7
Estes, Z., & Adelman, J. S. (2008). Automatic vigilance for negative words in lexical decision
and naming: Comment on Larsen, Mercer, and Balota (2006). Emotion, 8, 441–444;
discussion 445–457. doi:10.1037/1528-3542.8.4.441
Etkin, A., Egner, T., Peraza, D. M., Kandel, E. R., & Hirsch, J. (2006). Resolving emotional
conflict: A role for the rostral anterior cingulate cortex in modulating activity in the
amygdala. Neuron, 51, 871–882. doi:10.1016/j.neuron.2006.07.029
Ferré, P., Ventura, D., Comesaña, M., & Fraga, I. (2015). The role of emotionality in the
acquisition of new concrete and abstract words. Frontiers in Psychology, 6.
doi:10.3389/fpsyg.2015.00976
183
Fodor, J. A. (1975). The language of thought. Cambridge, MA: Harvard University Press.
Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT Press.
Fox, E., Russo, R., Bowles, R., & Dutton, K. (2001). Do threatening stimuli draw or hold visual
attention in subclinical anxiety? Journal of Experimental Psychology: General, 130, 681–
700.
Franz, M., Popp, K., Schaefer, R., Sitte, W., Schneider, C., Hardt, J., … Braehler, E. (2008).
Alexithymia in the German general population. Social Psychiatry and Psychiatric
Epidemiology, 43, 54–62. doi:10.1007/s00127-007-0265-1
Gardner, H. (1983). Frames of Mind: The Theory of Multiple Intelligences. Basic Book.
Gerhand, S., & Barry, C. (1999a). Age of acquisition, word frequency, and the role of phonology
in the lexical decision task. Memory & Cognition, 27, 592–602.
Gerhand, S., & Barry, C. (1999b). Age-of-acquisition and frequency effects in speeded word
naming. Cognition, 73, B27–36.
Gibbs, R. W. (2006). Metaphor Interpretation as Embodied Simulation. Mind & Language, 21,
434–458. doi:10.1111/j.1468-0017.2006.00285.x
Glenberg, A. M. (1997). What memory is for. The Behavioral and Brain Sciences, 20, 1–19;
discussion 19–55.
Glenberg, A. M. (2010). Embodiment as a unifying perspective for psychology. Wiley
Interdisciplinary Reviews: Cognitive Science, 1, 586–596. doi:10.1002/wcs.55
Glenberg, A. M., & Robertson, D. A. (2000). Symbol grounding and meaning: A comparison of
high-dimensional and embodied theories of meaning. Journal of Memory and Language,
43, 379–401.
Goerlich, K. S., Witteman, J., Aleman, A., & Martens, S. (2011). Hearing Feelings: Affective
184
Categorization of Music and Speech in Alexithymia, an ERP Study. PLoS ONE, 6,
e19501. doi:10.1371/journal.pone.0019501
Gollan, T. H., Slattery, T. J., Goldenberg, D., Van Assche, E., Duyck, W., & Rayner, K. (2011).
Frequency drives lexical access in reading but not in speaking: The frequency-lag
hypothesis. Journal of Experimental Psychology: General, 140, 186–209.
doi:10.1037/a0022256
Green, D. W. (2011). Language control in different contexts: The behavioral ecology of bilingual
speakers. Frontiers in Psychology: Language Sciences, 2, 103.
doi:10.3389/fpsyg.2011.00103
Grosjean, F. (2010). Bilingual: Life and reality. Cambridge, MA: Harvard University Press.
Harnad, S. (1990). The symbol grounding problem. Physica, 42, 335–346. doi:10.1016/01672789(90)90087-6
Harris, C. L., Ayçiçeĝi, A., & Gleason, J. B. (2003). Taboo words and reprimands elicit greater
autonomic reactivity in a first language than in a second language. Applied
Psycholinguistics, 24, 561–579. doi:10.1017/S0142716403000286
Harris, C. L., Gleason, J. B., & Ayçiçeği, A. (2006). When is a first language more emotional?
Psychophysiological evidence from bilingual speakers. In A. Pavlenko (Ed.), Bilingual
minds: Emotional experience, expression, and representation (pp. 257–283). Clevedon,
UK: Multilingual Matters.
Hauk, O., Johnsrude, I., & Pulvermüller, F. (2004). Somatotopic representation of action words
in human motor and premotor cortex. Neuron, 41, 301–307. doi:10.1016/S08966273(03)00838-9
Henderson, J. M., & Ferreira, F. (1993). Eye movement control during reading: Fixation
185
measures reflect foveal but not parafoveal processing difficulty. Canadian Journal of
Experimental Psychology, 47, 201–221.
Hino, Y., & Lupker, S. J. (1996). Effects of polysemy in lexical decision and naming: An
alternative to lexical access accounts. Journal of Experimental Psychology: Human
Perception and Performance, 22, 1331–1356. doi:10.1037/0096-1523.22.6.1331
Holt, D. J., Lynn, S. K., & Kuperberg, G. R. (2009). Neurophysiological correlates of
comprehending emotional meaning in context. Journal of Cognitive Neuroscience, 21,
2245–2262. doi:10.1162/jocn.2008.21151
Holt, D. J., Titone, D., Long, L. S., Goff, D. C., Cather, C., Rauch, S. L., … Kuperberg, G. R.
(2006). The misattribution of salience in delusional patients with schizophrenia.
Schizophrenia Research, 83, 247–256. doi:10.1016/j.schres.2005.12.858
Hyönä, J., & Häikiö, T. (2005). Is emotional content obtained from parafoveal words during
reading? An eye movement analysis. Scandinavian Journal of Psychology, 46, 475–483.
doi:10.1111/j.1467-9450.2005.00479.x
Inhoff, A. W., Starr, M., & Shindler, K. L. (2000). Is the processing of words during eye fixations
in reading strictly serial? Perception & Psychophysics, 62, 1474–1484.
James, C. T. (1975). The role of semantic information in lexical decisions. Journal of
Experimental Psychology: Human Perception and Performance, 1, 130–136.
doi:10.1037/0096-1523.1.2.130
Juhasz, B. J., Yap, M. J., Dicke, J., Taylor, S. C., & Gullick, M. M. (2011). Tangible words are
recognized faster: The grounding of meaning in sensory and perceptual systems.
Quarterly Journal of Experimental Psychology, 64, 1683–1691.
doi:10.1080/17470218.2011.605150
186
Kanske, P., & Kotz, S. A. (2007). Concreteness in emotional words: ERP evidence from a
hemifield study. Brain research, 1148, 138–148. doi:10.1016/j.brainres.2007.02.044
Kennedy, A., & Pynte, J. (2005). Parafoveal-on-foveal effects in normal reading. Vision
Research, 45, 153–168. doi:10.1016/j.visres.2004.07.037
Keuleers, E., Lacey, P., Rastle, K., & Brysbaert, M. (2011). The British Lexicon Project: Lexical
decision data for 28,730 monosyllabic and disyllabic English words. Behavior Research
Methods, 44, 287–304. doi:10.3758/s13428-011-0118-4
Kliegl, R., Nuthmann, A., & Engbert, R. (2006). Tracking the mind during reading: the influence
of past, present, and future words on fixation durations. Journal of Experimental
Psychology: General, 135, 12–35. doi:10.1037/0096-3445.135.1.12
Knickerbocker, H., Johnson, R. L., & Altarriba, J. (2015). Emotion effects during reading:
Influence of an emotion target word on eye movements and processing. Cognition &
Emotion, 29, 784–806. doi:10.1080/02699931.2014.938023
Kousta, S.-T., Vigliocco, G., Vinson, D. P., Andrews, M., & Del Campo, E. (2011). The
representation of abstract words: Why emotion matters. Journal of Experimental
Psychology: General, 140, 14–34. doi:10.1037/a0021446
Kousta, S.-T., Vinson, D. P., & Vigliocco, G. (2009). Emotion words, regardless of polarity, have
a processing advantage over neutral words. Cognition, 112, 473–481.
doi:10.1016/j.cognition.2009.06.007
Kroll, J. F., & Merves, J. S. (1986). Lexical access for concrete and abstract words. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 12, 92–107.
doi:10.1037/0278-7393.12.1.92
Kuchinke, L., Võ, M. L.-H., Hofmann, M., & Jacobs, A. M. (2007). Pupillary responses during
187
lexical decisions vary with word frequency but not emotional valence. International
Journal of Psychophysiology, 65, 132–140. doi:10.1016/j.ijpsycho.2007.04.004
Kuperman, V., Estes, Z., Brysbaert, M., & Warriner, A. B. (2014). Emotion and language:
Valence and arousal affect word recognition. Journal of Experimental Psychology:
General, 143, 1065–1081. doi:10.1037/a0035669
Lakoff, G., & Johnson, M. (1980). Metaphors We Live By. Chicago, IL: University of Chicago
Press.
Lang, P. J., & Bradley, M. M. (2010). Emotion and the motivational brain. Biological
Psychology, 84, 437–450. doi:10.1016/j.biopsycho.2009.10.007
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1990). Emotion, attention, and the startle reflex.
Psychological Review, 97, 377–395. doi:10.1037/0033-295X.97.3.377
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1997). Motivated attention: Affect, activation and
action. In P. J. Lang, R. F. Simons, & M. T. Balaban (Eds.), Attention and Orienting:
Sensory and Motivational Processes (pp. 97–135). Hillsdale, NJ: Lawrence Erlbaum
Associates, Inc.
Larsen, R. J., Mercer, K. A., & Balota, D. A. (2006). Lexical characteristics of words used in
emotional Stroop experiments. Emotion (Washington, D.C.), 6, 62–72. doi:10.1037/15283542.6.1.62
Larsen, R. J., Mercer, K. A., Balota, D. A., & Strube, M. J. (2008). Not all negative words slow
down lexical decision and naming speed: Importance of word arousal. Emotion, 8, 445–
452. doi:10.1037/1528-3542.8.4.445
Long, L. S., & Titone, D. A. (2007). Psychopathy and verbal emotion processing in nonincarcerated males. Cognition & Emotion, 21, 119–145.
188
doi:10.1080/02699930600551766
Louwerse, M. M., & Jeuniaux, P. (2008). Language comprehension is both embodied and
symbolic. In M. de Vega, A. M. Glenberg, & A. C. Graesser (Eds.), Symbols,
embodiment, and meaning (pp. 309–326). Oxford: Oxford University Press.
Louwerse, M. M., & Jeuniaux, P. (2010). The linguistic and embodied nature of conceptual
processing. Cognition, 114, 96–104. doi:10.1016/j.cognition.2009.09.002
Luminet, O., Vermeulen, N., Demaret, C., Taylor, G. J., & Bagby, R. M. (2006). Alexithymia and
levels of processing: Evidence for an overall deficit in remembering emotion words.
Journal of Research in Personality, 40, 713–733. doi:10.1016/j.jrp.2005.09.001
MacLeod, Tata, P., & Mathews, A. (1987). Perception of emotionally valenced information in
depression. British Journal of Clinical Psychology, 26 ( Pt 1), 67–68.
Marian, V., Blumenfeld, H. K., & Kaushanskaya, M. (2007). The language experience and
proficiency questionnaire (LEAP-Q): Assessing language profiles in bilinguals and
multilinguals. Journal of Speech, Language, and Hearing Research, 50, 940–967.
doi:10.1044/1092-4388(2007/067)
Matin, E. (1974). Saccadic suppression: A review and an analysis. Psychological Bulletin, 81,
899–917.
Matsumoto, D., Yoo, S. H., Hirayama, S., & Petrova, G. (2005). Development and validation of a
measure of display rule knowledge: The display rule assessment inventory. Emotion, 5,
23–40. doi:10.1037/1528-3542.5.1.23
McGinnies, E. (1949). Emotionality and perceptual defense. Psychological Review, 56, 244–251.
doi:10.1037/h0056508
McGrew, R. E. (1985). Encyclopedia of medical history. McGraw-Hill.
189
McKenna, F. P., & Sharma, D. (1995). Intrusive cognitions: An investigation of the emotional
Stroop task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21,
1595–1607. doi:10.1037/0278-7393.21.6.1595
Méndez-Bértolo, C., Pozo, M. A., & Hinojosa, J. A. (2011). Word frequency modulates the
processing of emotional words: Convergent behavioral and electrophysiological data.
Neuroscience Letters, 494, 250–254. doi:10.1016/j.neulet.2011.03.026
Monsell, S., Doyle, M. C., & Haggard, P. N. (1989). Effects of frequency on visual word
recognition tasks: Where are they? Journal of Experimental Psychology: General, 118,
43–71.
Mueller, J., Alpers, G. W., & Reim, N. (2006). Dissociation of rated emotional valence and
Stroop interference in observer-rated alexithymia. Journal of Psychosomatic Research,
61, 261–269. doi:10.1016/j.jpsychores.2006.02.017
Newcombe, P. I., Campbell, C., Siakaluk, P. D., & Pexman, P. M. (2012). Effects of emotional
and sensorimotor knowledge in semantic processing of concrete and abstract nouns.
Frontiers in Human Neuroscience, 6, 275. doi:10.3389/fnhum.2012.00275
Nyklicek, I., & Vingerhoets, A. (2000). Alexithymia is associated with low tolerance to
experimental painful stimulation. Pain, 85, 471–475. doi:10.1016/S03043959(99)00295-X
Paivio, A. (1965). Abstractness, imagery, and meaningfulness in paired-associate learning.
Journal of Verbal Learning and Verbal Behavior, 4, 32–38. doi:10.1016/S00225371(65)80064-0
Paivio, A. (1969). Mental imagery in associative learning and memory. Psychological Review,
76, 241–263. doi:10.1037/h0027272
190
Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart and Winston.
Pavlenko, A. (2005). Emotions and Multilingualism. Cambridge, UK: Cambridge University
Press.
Pavlenko, A. (2008). Emotion and emotion-laden words in the bilingual lexicon. Bilingualism:
Language and Cognition, 11, 147–164. doi:10.1017/S1366728908003283
Pavlenko, A. (2012). Affective processing in bilingual speakers: Disembodied cognition?
International Journal of Psychology, 47, 405–428. doi:10.1080/00207594.2012.743665
Pexman, P. M., Hargreaves, I. S., Edwards, J. D., Henry, L. C., & Goodyear, B. G. (2007). The
neural consequences of semantic richness: When more comes to mind, less activation is
observed. Psychological Science, 18, 401–406. doi:10.1111/j.1467-9280.2007.01913.x
Pexman, P. M., Lupker, S. J., & Hino, Y. (2002). The impact of feedback semantics in visual
word recognition: Number-of-features effects in lexical decision and naming tasks.
Psychonomic Bulletin & Review, 9, 542–549. doi:10.3758/BF03196311
Pollatsek, A., Juhasz, B. J., Reichle, E. D., Machacek, D., & Rayner, K. (2008). Immediate and
delayed effects of word frequency and word length on eye movements in reading: A
reversed delayed effect of word length. Journal of Experimental Psychology: Human
Perception and Performance, 34, 726–750. doi:10.1037/0096-1523.34.3.726
Ponari, M., Rodriguez-Cuadrado, S., Vinson, D., Fox, N., Costa, A., & Vigliocco, G. (2015).
Processing advantage for emotional words in bilingual speakers. Emotion.
Pratto, F., & John, O. P. (1991). Automatic vigilance: The attention-grabbing power of negative
social information. Journal of Personality and Social Psychology, 61, 380–391.
Pylyshyn, Z. (1999). Is vision continuous with cognition? The case for cognitive impenetrability
of visual perception. The Behavioral and Brain Sciences, 22, 341–365; discussion 366–
191
423.
Rayner, K. (1975). The perceptual span and peripheral cues in reading. Cognitive Psychology, 7,
65–81.
Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research.
Psychological Bulletin, 124, 372–422.
Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual search.
Quarterly Journal of Experimental Psychology, 62, 1457–1506.
doi:10.1080/17470210902816461
Rayner, K., & Duffy, S. A. (1986). Lexical complexity and fixation times in reading: Effects of
word frequency, verb complexity, and lexical ambiguity. Memory & Cognition, 14, 191–
201. doi:10.3758/BF03197692
R Development Core Team. (2010). R: A Language and environment for statistical computing.
Retrieved from http://www.R-project.org
Reichle, E. D., Pollatsek, A., & Rayner, K. (2006). E-Z Reader: A cognitive-control, serialattention model of eye-movement behavior during reading. Cognitive Systems Research,
7, 4–22. doi:10.1016/j.cogsys.2005.07.002
Reichle, E. D., Rayner, K., & Pollatsek, A. (2003). The E-Z reader model of eye-movement
control in reading: Comparisons to other models. The Behavioral and Brain Sciences, 26,
445–476; discussion 477–526.
Risse, S., & Kliegl, R. (2012). Evidence for delayed parafoveal-on-foveal effects from word n+2
in reading. Journal of Experimental Psychology. Human Perception and Performance,
38, 1026–1042. doi:10.1037/a0027735
Rizzolatti, G., & Craighero, L. (2004). The mirror-neuron system. Annual Review of
192
Neuroscience, 27, 169–192. doi:10.1146/annurev.neuro.27.070203.144230
Romani, C., McAlpine, S., & Martin, R. C. (2007). Concreteness effects in different tasks:
Implications for models of short-term memory. Quarterly Journal of Experimental
Psychology, 61, 292–323.
Russell, J. A. (2003). Core affect and the psychological construction of emotion. Psychological
Review, 110, 145–172.
Russell, J. A., & Barrett, L. F. (1999). Core affect, prototypical emotional episodes, and other
things called emotion: dissecting the elephant. Journal of Personality and Social
Psychology, 76, 805–819.
Schaefer, R., Schneider, C., Tress, W., & Franz, M. (2007). Cortical augmenting in alexithymic
subjects after unpleasant acoustic stimulation. Journal of Psychosomatic Research, 63,
357–364. doi:10.1016/j.jpsychores.2007.03.015
Schotter, E. R., Angele, B., & Rayner, K. (2012). Parafoveal processing in reading. Attention,
Perception & Psychophysics, 74, 5–35. doi:10.3758/s13414-011-0219-2
Schwanenflugel, P. J., Harnishfeger, K. K., & Stowe, R. W. (1988). Context availability and
lexical decisions for abstract and concrete words. Journal of Memory and Language, 27,
499–520. doi:10.1016/0749-596X(88)90022-8
Schwanenflugel, P. J., & Stowe, R. W. (1989). Context availability and the processing of abstract
and concrete words in sentences. Reading Research Quarterly, 24, 114–126.
doi:10.2307/748013
Scott, G. G., O’Donnell, P. J., Leuthold, H., & Sereno, S. C. (2009). Early emotion word
processing: Evidence from event-related potentials. Biological Psychology, 80, 95–104.
doi:10.1016/j.biopsycho.2008.03.010
193
Scott, G. G., O’Donnell, P. J., & Sereno, S. C. (2012). Emotion words affect eye fixations during
reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 38,
783–792. doi:10.1037/a0027209
Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3, 417–424.
doi:10.1017/S0140525X00005756
Segalowitz, N., Gatbonton, E., & Trofimovich, P. (2009). Links between ethnolinguistic
affiliation, self-related motivation and second language fluency: Are they mediated by
psycholinguistic variables? In Z. Dörnyei & E. Ushioda (Eds.), Motivation, language
identity and the L2 self (pp. 172–192). Clevedon: Multilingual Matters.
Segalowitz, N., Trofimovich, P., Gatbonton, E., & Sokolovskaya, A. (2008). Feeling affect in a
second language: The role of word recognition automaticity. The Mental Lexicon, 3, 47–
71. doi:10.1075/ml.3.1.05seg
Sheikh, N. A., & Titone, D. (2015a). The embodiment of emotional words in a second language:
An eye-movement study. Cognition & Emotion, 1–13.
doi:10.1080/02699931.2015.1018144
Sheikh, N. A., & Titone, D. (2015b). The parafoveal and foveal effects of negative and positive
emotional embodiment: Evidence form eye movements and the gaze-contingent boundary
paradigm. Submitted for publication to Emotion.
Sheikh, N. A., & Titone, D. A. (2013). Sensorimotor and linguistic information attenuate
emotional word processing benefits: An eye-movement study. Emotion, 13, 1107–1121.
doi:10.1037/a0032417
Sheppes, G., Scheibe, S., Suri, G., & Gross, J. J. (2011). Emotion-regulation choice.
Psychological Science, 22, 1391–1396. doi:10.1177/0956797611418350
194
Siakaluk, P. D., Pexman, P. M., Aguilera, L., Owen, W. J., & Sears, C. R. (2008). Evidence for
the activation of sensorimotor information during visual word recognition: The bodyobject interaction effect. Cognition, 106, 433–443. doi:10.1016/j.cognition.2006.12.011
Siakaluk, P. D., Pexman, P. M., Sears, C. R., Wilson, K., Locheed, K., & Owen, W. J. (2008).
The benefits of sensorimotor knowledge: Body-object interaction facilitates semantic
processing. Cognitive Science, 32, 591–605. doi:10.1080/03640210802035399
Sifneos, P. E. (1973). The prevalence of “alexithymic” characteristics in psychosomatic patients.
Psychotherapy and Psychosomatics, 22, 255–262. doi:10.1159/000286529
Sivik, T. (1993). Alexithymia and Hypersensitivity to Touch and Palpation. Integrative
Physiological and Behavioral Science, 28, 130–136. doi:10.1007/BF02691215
Sokal, M. M. (2001). Practical phrenology as psychological counseling in the 19th-century
United States. In C. D. Green, M. Shore, & T. Teo (Eds.), The transformation of
psychology: Influences of 19th-century philosophy, technology, and natural science (pp.
21–44). Washington, DC, US: American Psychological Association.
Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental
Psychology, 18, 643–662. doi:10.1037/h0054651
Sutton, T. M., Altarriba, J., Gianico, J. L., & Basnight-Brown, D. M. (2007). The automatic
access of emotion: Emotional Stroop effects in Spanish–English bilingual speakers.
Cognition & Emotion, 21, 1077–1090. doi:10.1080/02699930601054133
Talmy, L. (1983). How language structures space. In H. L. Pick & L. P. Acredolo (Eds.), Spatial
orientation: Theory, research, and application. New York: Plenum Press.
Taylor, S. E. (1991). Asymmetrical effects of positive and negative events: The mobilizationminimization hypothesis. Psychological Bulletin, 110, 67–85.
195
Tomasello, M. (2003). Constructing a language: A usage-based theory of language acquisition.
Cambridge, MA: Harvard University Press.
Tsai, J.-L., Kliegl, R., & Yan, M. (2012). Parafoveal semantic information extraction in
traditional Chinese reading. Acta Psychologica, 141, 17–23.
doi:10.1016/j.actpsy.2012.06.004
Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic
memory. Psychological Review, 80, 352–373. doi:10.1037/h0020071
Vigliocco, G., Kousta, S.-T., Della Rosa, P. A., Vinson, D. P., Tettamanti, M., Devlin, J. T., &
Cappa, S. F. (2014). The neural representation of abstract words: the role of emotion.
Cerebral Cortex (New York, N.Y.: 1991), 24, 1767–1777. doi:10.1093/cercor/bht025
Vigliocco, G., Meteyard, L., Andrews, M., & Kousta, S. (2009). Toward a theory of semantic
representation. Language and Cognition, 1, 219–248.
Vinson, D., Ponari, M., & Vigliocco, G. (2014). How does emotional content affect lexical
processing? Cognition & Emotion, 28, 737–746. doi:10.1080/02699931.2013.851068
Vuilleumier, P., & Huang, Y.-M. (2009). Emotional attention: Uncovering the mechanisms of
affective biases in perception. Current Directions in Psychological Science, 18, 148–152.
doi:10.1111/j.1467-8721.2009.01626.x
Vytal, K., & Hamann, S. (2009). Neuroimaging support for discrete neural correlates of basic
emotions: A voxel-based meta-analysis. Journal of Cognitive Neuroscience, 22, 2864–
2885. doi:10.1162/jocn.2009.21366
Wentura, D., Rothermund, K., & Bak, P. (2000). Automatic vigilance: the attention-grabbing
power of approach- and avoidance-related social information. Journal of Personality and
Social Psychology, 78, 1024–1037.
196
Whaley, C. P. (1978). Word—nonword classification time. Journal of Verbal Learning and
Verbal Behavior, 17, 143–154. doi:10.1016/S0022-5371(78)90110-X
Whitford, V., & Titone, D. (2012). Second-language experience modulates first- and secondlanguage word frequency effects: Evidence from eye movement measures of natural
paragraph reading. Psychonomic Bulletin & Review, 19, 73–80. doi:10.3758/s13423-0110179-5
Williams, J. M., Mathews, A., & MacLeod, C. (1996). The emotional Stroop task and
psychopathology. Psychological Bulletin, 120, 3–24.
Wilson, M. (2002). Six views of embodied cognition. Psychonomic Bulletin & Review, 9, 625–
636.
Wotschack, C., & Kliegl, R. (2013). Reading strategy modulates parafoveal-on-foveal effects in
sentence reading. The Quarterly Journal of Experimental Psychology, 66, 548–562.
doi:10.1080/17470218.2011.625094
Wundt, W. M. (1897). Outlines of Psychology. Retrieved from
http://psychclassics.yorku.ca/Wundt/Outlines/
Yang, J., Wang, S., Tong, X., & Rayner, K. (2010). Semantic and plausibility effects on preview
benefit during eye fixations in Chinese reading. Reading and Writing, 25, 1031–1052.
doi:10.1007/s11145-010-9281-8
Yan, M., Pan, J., Bélanger, N. N., & Shu, H. (2015). Chinese deaf readers have early access to
parafoveal semantics. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 41, 254–261. doi:10.1037/xlm0000035
Yan, M., Richter, E. M., Shu, H., & Kliegl, R. (2009). Readers of Chinese extract semantic
information from parafoveal words. Psychonomic Bulletin & Review, 16, 561–566.
197
doi:10.3758/PBR.16.3.561
Yan, M., & Sommer, W. (2015). Parafoveal-on-foveal effects of emotional word semantics in
reading Chinese sentences: Evidence from eye movements. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 41, 1237–1243. doi:10.1037/xlm0000095
Yan, M., Zhou, W., Shu, H., & Kliegl, R. (2012). Lexical and sublexical semantic preview
benefits in Chinese reading. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 38, 1069–1075. doi:10.1037/a0026935
Yap, M. J., & Balota, D. A. (2009). Visual Word Recognition of Multisyllabic Words. Journal of
Memory and Language, 60, 502–529. doi:10.1016/j.jml.2009.02.001
Yap, M. J., Tan, S. E., Pexman, P. M., & Hargreaves, I. S. (2011). Is more always better? Effects
of semantic richness on lexical decision, speeded pronunciation, and semantic
classification. Psychonomic Bulletin & Review, 18, 742–750. doi:10.3758/s13423-0110092-y
Zdrazilova, L., & Pexman, P. M. (2013). Grasping the invisible: Semantic processing of abstract
words. Psychonomic Bulletin & Review. doi:10.3758/s13423-013-0452-x
Zwaan, R. A. (2008). Experiential traces and mental simulations in language comprehension. In
M. DeVega, A. M. Glenberg, & A. C. Graesser (Eds.), Symbols, embodiment, and
meaning (pp. 165–180). Oxford: Oxford University Press.
Zwaan, R. A. (2014). Embodiment and language comprehension: Reframing the discussion.
Trends in Cognitive Sciences, 18, 229–234. doi:10.1016/j.tics.2014.02.008
198
APPENDICES
199
Appendix A
Sentences for Chapters 2 and 3
The valence condition is indicated for the target words within the three sentence frames for each
triplet by having them appear in the following order across the sentences:
negative/neutral/positive.
Tara read about the unbeliever/nightingale/benefactor that she overheard people discussing at
school.
Tina wrote about the unbeliever/nightingale/benefactor that the people on the bus were
discussing.
Tracy knew about the unbeliever/nightingale/benefactor that was now the topic of conversation.
The flyer was about the onslaught/fatherland/twilight that the group wanted to advertise.
The letter was about the onslaught/fatherland/twilight that was not being discussed overseas.
The pamphlet was about the onslaught/fatherland/twilight that had not been mentioned in the
media.
The editorial was about the weakness/subsection/blessing that other writers had also written
about.
The scrapbook was about the weakness/subsection/blessing that the newspapers had frequently
written about.
The website was about the weakness/subsection/blessing that the principal spoke of at the
meeting.
Albert said that the farewell/tablespoon/painting that he saw was ordinary.
Andrew felt that the farewell/tablespoon/painting that he received was adequate.
Arthur knew that the farewell/tablespoon/painting that he gave was sufficient.
The drawing included the sickness/typewriter/sunlight that the artist spoke of during the
interview.
The painting depicted the sickness/typewriter/sunlight that was out on the street.
The picture captured the sickness/typewriter/sunlight that influenced the author's artistic choices.
Marianne was thinking about the blunder/dictation/slumber that the school committee was
discussing in their meeting.
Marlene was talking about the blunder/dictation/slumber that her son's teacher brought up during
the meeting.
Monique was writing about the blunder/dictation/slumber that the education expert spoke about
during the presentation.
The guests chatted about the failure/beginning/welcome that the host told them about.
The members emailed about the failure/beginning/welcome that had raised concerns at the last
meeting.
The neighbors talked about the failure/beginning/welcome that they read about in a magazine.
200
Alison spoke about the disease/statement/evening that was covered on an episode of 60 Minutes.
Annie read about the disease/statement/evening that was described in the morning paper.
Ashley wrote about the disease/statement/evening that she read about in the Washington Post.
The couple contemplated the tobacco/vegetable/bedroom that they needed tomorrow evening.
The parents considered the tobacco/vegetable/bedroom that their son had requested.
The roommates discussed the tobacco/vegetable/bedroom that they would prepare for the party.
Felix remembered the worm/rug/gig that he read about in the paper.
Frank photographed the worm/rug/gig that was shown at the festival.
Fred described the worm/rug/gig that he had seen while traveling.
The officer spoke about the horn/eye/girl that his colleague heard about at the station.
The student heard about the horn/eye/girl that his fellow students discussed in class.
The teacher wrote about the horn/eye/girl that was described in the manuscript.
Lauren joked about the lie/tip/ride that her roommate had kept secret.
Linda talked about the lie/tip/ride that her colleague had shared with her.
Lisa gossiped about the lie/tip/ride that her friends had spoken of.
Martin heard about the jerk/elm/joke that everyone was talking about.
Matthew spoke about the jerk/elm/joke that he read about at school.
Michael wrote about the jerk/elm/joke that was well known in his hometown.
The essay mentioned the hell/call/love that changed the author's life.
The paper emphasized the hell/call/love that precipitated the character's downfall.
The story underscored the hell/call/love that symbolized the protagonist's humanity.
The contract was about the drain/seat/shore that needed to be maintained.
The meeting was about the drain/seat/shore that required cleaning.
The message was about the drain/seat/shore that was ignored over the winter.
Mandy noticed that the blood/book/water that she brought up seemed to have ended the
conversation.
Marie doubted that the blood/book/water that she wrote about was worth including in the
summary.
Mary concluded that the blood/book/water that she heard about should be included in the report.
The audience heard about the harm/fate/pride that would follow the election.
The parents spoke about the harm/fate/pride that followed their child's choice.
The reporters wrote about the harm/fate/pride that would follow the nomination.
Jack approached the stump/leek/otter that he came upon on his way to school.
James avoided the stump/leek/otter that he saw on his way to the river bank.
Jeff described the stump/leek/otter that he saw in his backyard.
201
Barbara read about the scare/tale/lunch that the airplane crew was talking about.
Brenda spoke about the scare/tale/lunch that became the sole focus of the conversation.
Bridget heard about the scare/tale/lunch that people had been gossiping about.
The reporter covered the punch/root/honey that had made the headlines recently.
The student discussed the punch/root/honey that was described in last week's reading.
The teacher included the punch/root/honey that was reported by the media in her lecture.
The art teacher presented the smoke/hill/drink that the students were going to paint.
The news report mentioned the smoke/hill/drink that was discussed at work.
The school paper reviewed the smoke/hill/drink that had caused concern among students.
Richard read about the slime/seal/novel that was mentioned in the papers.
Robert wrote about the slime/seal/novel that he saw on television.
Ronald spoke about the slime/seal/novel that his wife showed him on the internet.
They heard about the crook/cork/jewel that was written about in the paper.
They spoke about the crook/cork/jewel that they learned about at school.
They wrote about the crook/cork/jewel that they read about on a website.
Adrian heard about the greed/mood/youth that was central to the election campaign.
Andrew spoke about the greed/mood/youth that the city policy was aimed at.
Anthony read about the greed/mood/youth that the council was concerned about.
The children inquired about the flood/belt/heart that was discussed on television.
The couple wondered about the flood/belt/heart that was shown in the commercial.
The customers asked about the flood/belt/heart that was mentioned in an announcement.
They heard about the curse/poll/movie that was ridiculed by the critics.
They spoke about the curse/poll/movie that was popular across the nation.
They wrote about the curse/poll/movie that was ignored by the government.
The email was about the brawl/hail/spring that the municipal government was planning for.
The exchange was about the brawl/hail/spring that was described in the community paper.
The message was about the brawl/hail/spring that was made fun of in the political cartoon.
The magazine article was about the filth/wing/cheese that was now a popular topic of
conversation.
The science documentary was about the filth/wing/cheese that was also written about in the New
York Times.
The television show was about the filth/wing/cheese that was included in today's lecture at
school.
Daniel was writing about the tomb/ton/ball that was discussed after last night's hockey game.
David was asking about the tomb/ton/ball that his colleague had brought up over lunch
202
yesterday.
Derek was thinking about the tomb/ton/ball that he had seen on the Discovery Channel last night.
Sandy wrote about the gas/bay/lake that she had learned about from her science teacher.
Sarah read about the gas/bay/lake that was described in her science textbook.
Sylvia spoke about the gas/bay/lake that has recently been discussed at length in the media.
The developer designed the cane/core/park that was at the heart of a large development contract.
The engineer examined the cane/core/park that she was going to study as part of her training.
The officer investigated the cane/core/park that was at the center of the shopping center.
The brothers spoke about the fever/boot/crown that they discussed earlier during breakfast.
The friends read about the fever/boot/crown that was described in an article in the municipal
paper.
The students wrote about the fever/boot/crown that their teacher described during a lecture.
The phone call was about the debt/act/sex that the two writers were going to work into their
movie script.
The text message was about the debt/act/sex that the two friends were going to talk about over
coffee.
The voice mail was about the debt/act/sex that had been discussed last night after a few drinks at
the bar.
Edward described the fury/vow/hero that he had read about in the science fiction novel.
Elliot discussed the fury/vow/hero that he was going to include in his story about ancient Rome.
Eric remembered the fury/vow/hero that was shown in the play that he attended last week.
The art student spoke about the punishment/temptation/excitement that he had to capture in his
drawing assignment.
The law student wrote about the punishment/temptation/excitement that was demonstrated in the
police video.
The school girl read about the punishment/temptation/excitement that was described in her
assignment.
Maria wrote about the siren/cave/feast that she dreamed about after meeting her new boyfriend.
Melinda spoke about the feast she had seen the night before in her surreal painting.
Melinda spoke about the siren/cave/feast that she had seen the night before in her surreal
painting.
The business student read about the panic/risk/glory that was described in his text on economic
successes and failures.
The English student spoke about the panic/risk/glory that was illustrated in the Shakespearean
play.
The history student wrote about the panic/risk/glory that was discussed in Machiavelli's treatise
on warfare.
203
She read about the cynic/cord/angel that was described in the book that her friend had lent to her.
She spoke about the cynic/cord/angel that was shown on the popular television show last night.
She wrote about the cynic/cord/angel that she had dreamed about last week after her midterm.
Randy spoke about the demon/bass/charm that he thought characterized the statesman's voice
during the speech.
Richard read about the demon/bass/charm that an author had used to make his novel's character
seem unusual.
Robert wrote about the demon/bass/charm that his wife had described after seeing it on a soap
opera.
The new student contemplated the government/character/experience that was part of the
scholarly organization.
The old lawyer described the government/character/experience that was part of the international
conference.
The young artist depicted the government/character/experience that was part of the overseas
exhibition.
Ally was writing about the forfeit/sedative/comrade that she described in her memoirs.
Anna was thinking about the forfeit/sedative/comrade that she remembered when reminiscing
about high school.
Ava was speaking about the forfeit/sedative/comrade that her new husband had just learned about
the night before.
The farmer remembered the gang/mile/hobby that his colleagues were just inquiring about.
The teacher described the gang/mile/hobby that was made famous by a series of pictures.
The writer recalled the gang/mile/hobby that his mentor often depicted in his own books.
Ben fell on the brute/dame/surf that he came across while not paying attention.
Bill went after the brute/dame/surf that he saw while walking on the beach.
Bob ran toward the brute/dame/surf that he caught a glimpse of from the boardwalk.
Jake mentioned the prison/face/film that they disliked for various reasons.
Jeff ignored the prison/face/film that was written about in the magazine.
Josh inspected the prison/face/film that everyone had been talking about.
The lecture was about the poison/suite/justice that was discussed in the first chapter.
The paper was about the poison/suite/justice that was discussed in the first act.
The question was about the poison/suite/justice that was discussed in the first half of the book.
Travis spoke about the destruction/investigation/imagination that he was planning to talk about at
work.
Trevor wrote about the destruction/investigation/imagination that was discussed during the
conference call.
Truman read about the destruction/investigation/imagination that was described in an issue of
People Magazine.
204
Tammy commented on the interruption/imperfection/compliment that had provoked an odd
reaction from her husband.
Tanya responded to the interruption/imperfection/compliment that she had encountered while
surfing the internet.
Taylor reacted to the interruption/imperfection/compliment that had garnered so much attention
around the office.
The drawing was of the obstruction/inauguration/innocence that was also depicted on the front
cover of the book.
The painting was of the obstruction/inauguration/innocence that was also illustrated in the
magazine.
The sketch was of the obstruction/inauguration/innocence that was also drawn on the back cover
of the novel.
The conference was about the suspect/route/revenue that was discussed in the media.
The conversation was about the suspect/route/revenue that was discussed in class.
The presentation was about the suspect/route/revenue that was discussed in the paper.
Mary read about the division/cause/concert that was going to transpire in the following year.
Maya spoke about the division/cause/concert that was bound to create a large sum of publicity.
Megan wrote about the division/cause/concert that was being planned by the group of partners.
Martin discussed the impatience/commencement/affection that had triggered the whole chain of
events.
Michael documented the impatience/commencement/affection that he witnessed between his two
close friends.
Milton remembered the impatience/commencement/affection that he saw at his last place of
business.
Barney read about the mutilation/intoxication/gentleman that influenced the lawmakers to sign
the new bill.
Bobby wrote about the mutilation/intoxication/gentleman that inspired the movie director to
produce a film.
Bradley spoke about the mutilation/intoxication/gentleman that caused the increase in television
viewers.
205
Appendix B
Supplementary Material For Chapter 3
L2 Readers’ Background
We analyzed the Language Experience and Proficiency Questionnaire (LEAP-Q; Marian,
Blumenfeld, & Kaushanskaya, 2007) data to confirm that the L2 readers were less proficient in
their L2 than L1, and primarily used their L2 in formal environments, rather than informal social
environments. We used self-report to measure language proficiency and use of the L2 in
different contexts (both using 7-point scales) and the percentage of time exposed to the L2 (see
Table 1).
206
Table 1. Measures For Language Proficiency, Exposure, And Use In Different Contexts
Measure
Variable
Range
L1 proficiency
7-point scale 1 (beginner) to 7 (near-native)
L2 proficiency
7-point scale 1 (beginner) to 7 (near-native)
L2 exposure at the time of the experiment Percentage
0-100%, but must sum to 100 when combined with L1 exposure
L2 use at home
7-point scale 1 (none at all) to 7 (a significant amount)
L2 use at work
7-point scale 1 (none at all) to 7 (a significant amount)
L2 use in social situations
7-point scale 1 (none at all) to 7 (a significant amount)
207
We analyzed relationships between these measures of proficiency, exposure, and use of
the L2 in different contexts. First, we confirmed that the participants were less proficient in their
L2 (M = 6, SD = 1.08) compared to their L1 (M = 7, SD = 0.53). Next, we used linear regression
to examine how L2 proficiency predicted L2 exposure and L1 proficiency. We fit a model to the
L2 proficiency data and specified main effects for L2 exposure and L1 proficiency. Higher L2
exposure (b = 0.04, SE = 0.01, t = 2.75, p < 0.01) and L1 proficiency (b = 0.78, SE = 0.40, t =
1.96, p = 0.059) predicted higher L2 proficiency after controlling for each variable. We then
examined how L2 exposure was correlated with L2 usage across home use, social use, and work
use. We fit a model to the L2 exposure data and specified main effects for home use, social, use,
and work use of the L2. Only work use of the L2 predicted L2 exposure (b = 3.00, SE = 1.29, t =
2.33, p < 0.05). Thus, L2 proficiency was primarily associated with L2 exposure in formal
environments in our participants.
Fixation Probability & Regression Probability
We modeled fixation probability and regression probability to supplement the main
analysis of fixation duration measures. For fixation probability, there was an interaction between
positive valence and frequency (b = -0.33, SE = 0.15, z = -2.23, p < 0.05) and an interaction
between negative valence and concreteness (b = -0.38, SE = 0.13, z = -2.88, p < 0.01). A model
with negative valence as the baseline showed that the valence by frequency interaction also
differed for negative vs. positive words (b = -0.35, SE = 0.17, z = -2.06, p < 0.05) and was
trending for valence by concreteness (b = 0.25, SE = 0.15, z = 1.70, p = 0.09). To follow up on
these interactions, we fit models to the data split (separately) by valence, frequency, and
concreteness, which showed that fixation probability was higher for low-frequency positive
words than high-frequency positive words (b = -0.40, SE = 0.13, z = -3.06, p < 0.01), higher for
abstract negative words than concrete negative words (b = -0.23, SE = 0.10, z = -2.36, p < 0.05),
higher for low-frequency positive words than low-frequency negative words (b = 0.54, SE =
208
0.22, z = 2.48, p < 0.05), and trending to be higher for low-frequency positive words than lowfrequency neutral words (b = 0.40, SE = 0.23, z = 1.77, p = 0.08). No other effect for valence
was significant and there were no interactions involving proficiency.
For regression probability, there was a main effect of negative valence indicating that
regression probability was lower for negative words than neutral words (b = -0.27, SE = 0.14, z =
-1.93, p = 0.05). There was also an interaction between positive valence and frequency (b =
0.37, SE = 0.13, z = 2.88, p < 0.01) and an interaction between positive valence and concreteness
(b = -0.26, SE = 0.13, z = -2.03, p < 0.05). A model with negative valence as the baseline
showed that the valence by frequency interaction and the valence by concreteness interaction
were also significant for negative vs. positive words (b = 0.43, SE = 0.15, z = 2.80, p < 0.01; b =
-0.36, SE = 0.14, z = -2.55, p < 0.05; respectively). To follow up on these interactions, we fit
models to the data split (separately) by valence, frequency, and concreteness, which showed that
regression probability was higher for high-frequency positive words than low-frequency positive
words (b = 0.43, SE = 0.13, z = 3.33, p < 0.001), and trending to be lower for low-frequency
positive words than low-frequency neutral words (b = -0.32, SE = 0.19, z = -1.69, p = 0.09), and
trending to be higher for high-frequency positive words than high-frequency negative words (b =
0.41, SE = 0.22, z = 1.86, p = 0.06). Trends suggest that the effects underlying the interaction
between positive valence and concreteness were higher regression probability for concrete
neutral words than concrete positive words (b = -0.30, SE = 0.20, z = -1.54, p = 0.13) and higher
regression probability for concrete negative words than abstract negative words (b = 0.22, SE =
0.14, z = 1.59, p = 0.11). No other effect for valence was significant and there were no
interactions involving proficiency.
209