“Micro-valences”: Affective valence in “neutral”...
Transcript of “Micro-valences”: Affective valence in “neutral”...
“Micro-valences”: Affective valence in “neutral”
everyday objects
By Sophie Lebrecht M.A., University of Glasgow, 2005
A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN THE
DEPARTMENT OF COGNITIVE, LINGUSITIC, AND PSYCHOLOGICAL SCIENCES AT BROWN UNIVERSITY
PROVIDENCE, RHODE ISLAND MAY 2012
© Copyright 2012 by Sophie Lebrecht
iii
This dissertation by Sophie Lebrecht is accepted in its present form by the Department of Cognitive, Linguistic and Psychological Sciences as satisfying the dissertation
requirement for the degree of Doctor of Philosophy.
Date David L. Sheinberg, Ph.D., Director
Recommended to the Graduate Council Date: Michael J. Tarr, Ph.D., Reader Date: David Badre, Ph.D., Reader Date: Moshe Bar, Ph.D., Reader
Approved by the Graduate Council
Date: Peter M. Weber, Dean of the Graduate School
iv
SOPHIE LEBRECHT
Department of Cognitive and Linguistic Sciences 116 Governor Street Brown University Providence, RI 02906 Box: G-LN Tel: (401) 863-6876 Cell: (401) 533-7384 http://www.cnbc.cmu.edu/~lebrecht http://tarrlab.cnbc.cmu.edu/ Citizenship: UK Email: [email protected] Permanent Resident Status of United Status EDUCATION
Brown University
Department of Cognitive and Linguistic Sciences Ph.D. student, Expected graduation date 2011
University of Glasgow
Department of Psychology Received Master of Arts, July 2005
University of California, San Diego Department of Psychology Junior Year Abroad, 2004-05
HONORS AND AWARDS 2004-05 Awarded Full Tuition and Stipend, International Exchange Program, UCSD-
Glasgow Exchange Program 2005 First Class Honors, University of Glasgow 2007 NEI Vision Training Grant, Full Tuition and Stipend for 3 years 2009 International Graduate Student Travel Award 2009 Golden Greeble Grant from Perceptual Expertise Network 2010 Trainee Grant from Temporal Dynamics of Learning Centre 2010 International Graduate Student Travel Award 2010 AAUW Dissertation Fellowship
RESEARCH EXPERIENCE 2005 Independent Study with Philippe Schyns, Facial Categorization 2004–05 Honors Thesis with David Simmons, The Binocular Summation of color and
Luminance Contrast 2003–04 Research Assistant for Stuart Anstis, Color and Motion Adaptation Effects
TEACHING / GUEST LECTURES
2007 Face Perception and the Brain, Undergraduate Lecture, RISD 2008 Theatre and Neuroscience, Teaching Assistant 2009 Perception and the Mind, Teaching Assistant
v
2009 Functional Magnetic Resonance Imaging: Theory and Practice, Teaching Assistant 2009 How Implicit Biases affect our Teaching Practices, Graduate workshop hosted by
the Sheridan Centre for Teaching, Brown University 2010 How Implicit Biases affect our Teaching Practices, Faculty workshop hosted by
the Sheridan Centre for Teaching, Brown University 2010 Introduction to Cognitive Neuroscience, Teaching Assistant 2010 Brown Ethical and Responsible Conduct of Research Education (BEARCORE)
Program Undergraduate Honors Supervision 2008 – 2010 Leslie Roos: Awarded National Science Foundation Travel Award through the Temporal Dynamics of Learning Centre (2009) & Dept. of Psychology prize for exceptional honors thesis. 2008 – 2010 David Pagliaccio: Awarded Brown Undergraduate Research Training Award (2009) & Department of Neuroscience prize for exceptional honors thesis.
RESEARCH INTERESTS Visual Perception, Face and Object Recognition, understanding the relation of Visual Perception to Semantic and Affective Cognitive Neuroscience. DISSERTATION COMMITTEE David Badre (Brown University), Moshe Bar (Harvard Medical School, MGH), David Sheinberg (Brown University), Michael J. Tarr (Carnegie Mellon University). PUBLICATIONS Lebrecht, S., Johnson, D. M., & Tarr, M. J. Submitted. The Affective Lexical Priming Score. Roos, L., Lebrecht, S., Tanaka, J. W., & Tarr, M. J. Under Review. Can we change implicit racial biases? “Yes we can!” Lebrecht, S. & Tarr, M. J. Blindsight. Clinical Neuropsychology Encyclopedia. In press. Lebrecht, S. & Tarr, M. J. Achromatopsia. Clinical Neuropsychology Encyclopedia. In press. McGugin, R. W., Tanaka, J. W., Lebrecht, S., Tarr, M. J., & Gauthier, I. (2010). Race-Specific perceptual discrimination improvement following short individuation training with faces. Cognitive Science, 35, 330-347. Lebrecht S, Pierce LJ, Tarr MJ, Tanaka JW. (2009). Perceptual Other-Race Training Reduces Implicit Racial Bias. PLoS ONE 4(1): e4215. doi:10.1371/journal.pone.0004215 Lebrecht, S., & Badre, D. (2008). Emotional regulation, or: How I learned to stop worrying and love the nucleus accumbens. Neuron, 59(6), 841-3.
vi
CONFERENCE PRESENTATIONS Symposium Lebrecht S, Pierce L.J, Tarr M.J, Tanaka J.W. Perceptual Other Race Training Reduces Implicit Racial Bias (2009). Applied Psychological Society Annual Meeting. Talks Interactions Between Visual and Other Cognitive Systems (2010). Department of Psychology/Institute for Neuroscience, Newcastle University, Colloquia series. Perceptual Other Race Training & Implicit Racial Bias (2009). University of Hong Kong, Vision Group Seminar Series. Controlling Semantic and Emotional Associations in the Brain: A Proposed Study (2008). Perceptual Expertise Network Meeting, Banff. The Other Race Effect: A Predictor of Social Bias? (2007). Perceptual Expertise Network Meeting, Boston. Posters Lebrecht, S., Roos, L.E., Bar, M., & Tarr, M.J. (2010). Are everyday objects ever affectively “neutral”? Temporal Dynamics of Learning Center: Project 2.3.3 (NSF). Lebrecht, S. & Tarr, M. J. (2010). Defining an Object’s Micro-Valence through Implicit Measures. Graduate NSF Institute for Science Learning Conference. McGugin, R., Tanaka J.W, Lebrecht, S., Tarr, M.J., & Gauthier, I. (2010). Race-specific perceptual discrimination improvement following short individuation training with faces. [Abstract]. Journal of Vision, 10(7): 622. Lebrecht, S. & Tarr, M. J. (2010). Defining an Object’s Micro-Valence through Implicit Measures [Abstract]. Journal of Vision, 10(7): 966. Pagliaccio, D., Lebrecht, S., & Badre, D. (2010). Distributed Patterns of Activation in the Pre-frontal Cortex Reflect Task Relevant Visual Dimensions of a Stimulus. Society of Cognitive Neuroscience. Lebrecht, S., Pagliaccio, D., Long, M., & Badre, D. (2009). Ventrolateral Prefrontal Cortex Contributions to Rule Guided Memory Retrieval. Society for Neuroscience Annual Meeting. Lebrecht, S., Righi, G., Vettel, J. M., & Tarr, M. J. (2008). Predicting Individual Differences in fMRI Cortical Activation using Behavioral Measures. Society for Neuroscience Annual Meeting. Lebrecht, S., Pierce, L., Tanaka, J., & Tarr, M. J. (2008). Seeing Beyond Faces: The Social Significance of Being an Other-Race Expert [Abstract]. Journal of Vision, 8(6): 259, 259a.
vii
MEMBERSHIPS Society for Neuroscience Vision Sciences Society Applied Psychology Society National Science Foundation: Temporal Dynamics of Learning Centre Trainee Perceptual Expertise Network Trainee JOURNAL REFREE Neuron Journal of Experimental Psychology Letters to Neuroscience PLoS One
viii
Acknowledgements Many people contributed their time, effort and ideas in completion of this
dissertation. I thoroughly enjoyed working on my PhD, which is without question due to the inspiring, supportive, down-to-earth scientists with whom I had the fortune of working. In particular, I would like to thank my entire committee, who has each contributed to my training and dissertation in their own way.
Michael J. Tarr gave me the freedom and confidence to pursue ideas that I truly believed in.
David Badre taught me the importance of accuracy, precision, and control in my experiments. He devoted the time and the care to teach me the complexities of neuroimaging, imbuing in me a deeper understanding that I continually refer to as I develop my own work.
Moshe Bar inspired me with such creativity and imagination in his approach to perception. He always encouraged me to take a novel perspective when devising theories and interpreting data. I will always be grateful for the warmth and openness with which he welcomed me into his lab.
David Sheinberg offered me continued and upbeat motivation that kept me going during the harder times of completing my thesis. He has guided me throughout my PhD, but the last two years in particular his support and direction has been exceptional. David has happily worked on the administrative side of having a student, for which I am truly grateful. David’s comments on my work have always been some of the most insightful, even when they are delivered as he is flying off down the corridor! I am deeply grateful for his commitment to myself and my research and have enjoyed every moment of being part of his group.
I would like to add a special thank you to my advisor, Mike Tarr. As an advisor, Mike always listened carefully to my ideas and valued my opinions and contributions to our work. He trusted the directions that I wanted to pursue and provided me with countless opportunities to create novel and exciting science. He always believed in my work and had confidence in my ability as a scientist (even when I doubted myself). This unwavering support and commitment has given me the skills, opinions, and confidence to enter the field as an independent scientist. I feel incredibly grateful to have had the privilege of being mentored by Mike, and I am sure all I have learnt from him will continue to guide me throughout my career.
I would also like to express my appreciation to all of the people who helped me get to this point: Deborah M. Johnson, David Munoz, Darren Seibert, Leslie Roos, Lynn Fanella, Erika Nixon, Mike Worden, Ed Walsh, Denise Camera, David Pagliaccio, Nicole Long, Lisa Barrett, James Tanaka, Lisa Egan, E. T. Cunningham, Marlene Behrmann, David Plaut, Yasmine Kahn, The Brown University Writing Center. I would like to thank all of the members (past and present) of the Tarr Lab, Sheinberg Lab, Badre Lab, and the Bar Lab, and in particular to Jean Vettel, Adrian Nestor, Giulia Righi, Ilke Oztekin, Heida Sigurdardottir, and Luke Woloszyn. I extend a huge thank you to my friends and family, for their patience and encouragement. And to my husband, Jason Groves, whose contribution has been so vast and intimate that I cannot reduce it down to words. Just to say, thank you Jason, I honestly couldn’t have kept going without you.
ix
Table of contents
1. “Micro-valence”: Affective valence in everyday objects....................1
1.1 Introduction ...........................................................................................1
1.2 Object valence operates on a continuum ..............................................4
1.3 Understanding the origins of micro-valence..........................................7
1.4 The perceptual components of micro-valence ......................................10
1.5 How the perception of valence relates to the discourse of aesthetics..11
1.6 Object valence and preference..............................................................12
1.7 Clinical implications for object valence..................................................14
1.8 Why have the affective properties of everyday objects been
underestimated? ..........................................................................................16
1.9 Discussion..............................................................................................17
2. The Affective Lexical Priming Score ....................................................19
2.1 Object norming ......................................................................................24
2.2 Word norming ........................................................................................27
2.3 The Affective Lexical Priming Score ......................................................32
3. Perceiving valence in everyday objects...............................................42
3.1 The “birthday” task ................................................................................45
3.2 The ranking task.....................................................................................50
3.3 How micro-valence influences object recognition.................................54
4. The neural basis of object recognition ................................................63
5. Discussion ..............................................................................................81
5.1 What is micro-valence? .........................................................................81
5.2 Is micro-valence micro-affective?..........................................................86
5.3 Micro-valence and preference ...............................................................87
5.4 Model for object valence .......................................................................88
x
List of tables
2.1 Mean (M) Reaction Times (RT) and Standard Deviations (SD) for positive, negative, and neutral words.........................................................................31
2.2 Mean (M) Reaction Times (RT) and Standard Deviations (SD) for positive, negative, and neutral words in final word set ..............................................32
3.1 Proportion of participants who perceived micro-valence objects with high consistency ..........................................................................................49
4.1 MNI coordinates for the peak voxels in the strong and micro-valence conditions for individual participants ...........................................................73
xi
List of figures
1.1 The valence continuum .........................................................................5
2.1 Experimental design (ALPS)...................................................................35
2.2 Reaction times for congruent and incongruent valence conditions ......36
2.3 Reaction times for all conditions in ALPS..............................................37
3.1 Consensus in the perception of micro-valence .....................................48
3.2 Correlation between valence scores from the “birthday” task and the
ranking task..................................................................................................52
3.3 Heatmap for individual participant correlations between valence scores
from the “birthday” task and ranking task ..................................................53
3.4 Response times for identity judgments in a matching task ...................58
3.5 Valence as a feature of objects..............................................................61
4.1 Response times for pleasantness judgments........................................71
4.2 Activation map for strong and micro-valence........................................72
4.3 Activation map for the center of the valence continuum .......................74
4.4 Time course of valence-related BOLD signal in prefrontal cortex .........75
4.5 Relationship between BOLD signal and valence ...................................76
4.6 Valence processing in the lateral occipital cortex .................................77
5.1 Valence ratings for novel shapes ...........................................................84
5.2 Difference in valence for high quality versus low quality images...........85
5.3 A model for object valence ....................................................................89
1
1. INTRODUCTION
“Micro-valence”: Affective valence in everyday objects
1.1 Introduction
While grabbing a coffee mug from the cupboard, the phone rings, your attention
is diverted and, in the split second before answering the call, you select a particular
mug from your collection. What factors guide you to choose one mug over the others?
Here we aim to answer this question by proposing that the majority of objects around
us possess some valence – “micro-valence” – which ranges in magnitude but is almost
always present. The notion of micro-valence has direct implications for the perception
of and interaction with our surroundings in that our argument is not limited only to
objects with strong valence (e.g. guns or roses). Rather, we suggest that all everyday
objects – even seemingly neutral ones – evoke some perception of valence. This
explanation of micro-valence illustrates a larger point – that visual object perception is
linked to affective processing via mechanisms that automatically evaluate and assign
valences to objects that then interact with our perceptual, cognitive, social, and
affective systems.
The vast majority of research on affective object perception focuses on objects
that generate a strong, well-defined valence (Colibazzi et al., 2010; Rudrauf et al., 2008;
Weierich, Wright, Negreira, Dickerson, & Barrett, 2010); however, the extent to which
everyday objects (e.g., lamps or clocks) generate more subtle valences has yet to be
examined in detail. Typically, affect is described in terms of two continuous dimensions:
valence (pleasantness) and arousal (activation) (Barrett, 2006; Russell, 1980; Russell &
Carroll, 1999). However, with respect to micro-valences we are primarily focused on the
single dimension of valence.
2
Even though, for purposes of motivating the idea of micro-valence, we posit that
all objects are coded along a single valence dimension or continuum, we do not want to
downplay the notable differences between the response to a bloody weapon and a
coffee mug, namely the intensity of and the variance in one’s responses. A bloody
weapon will generate a highly intense affective response and, at the same time, the
between-subject variance in response should be relatively low. That is, the vast majority
of us will experience the same highly negative feeling when viewing it.
On the other hand, the coffee mug will likely generate a weaker or subtler
response, which we refer to as an object’s “micro-valence.” This valence is described
as “micro”, because the intensity of the response is less than the bloody weapon and
other similarly strongly affective objects. At the same time, this weak intensity should
not be confused with a weak effect. In science there are many small, but robust effects.
For example, Sternberg’s (1966) classic digit memory search exhibited an effect of less
than 40 msec per and item in memory (Sternberg, 1966), and functional magnetic
resonance imaging (fMRI) typically relies on a signal change of only between 0.5 –
2.0%.
What then can the concept of micro-valence do to help inform the question:
how do you select a mug from a cupboard full of equally suitable mugs? Our
hypothesis is that during visual object perception the valences of objects are
automatically computed. Moreover, nominally neutral objects such as mugs carry
micro-valences, and, as such, we predict that you are slightly more likely to pick a mug
with a positive micro-valence and avoid a mug with a negative micro-valence. This
happens because, once computed, these valences provide input to a network of
valuation and decision-making mechanisms (Grabenhorst & Rolls, 2011; Rangel,
Camerer, & Montague, 2008; Sugrue, Corrado, & Newsome, 2005).
3
Given our assumption that the evaluation of micro-valence occurs during object
perception on the order of milliseconds, it follows that observers preferentially select
certain mugs quickly and without conscious evaluation. Similarly, micro-valences may
also influence slower, more deliberative decisions, but critically the valence information
generated during perception is available almost simultaneously with visual recognition
and can therefore be used at any subsequent point during decision-making or behavior.
Consider for a moment a completely neutral world, devoid of valence. What
would happen if observers did not generate valences during perception? We suggest
that this is a reality for the neuropsychological patient, EVR, who suffered damage to
his orbitofrontal cortex related to the removal of a tumor (Eslinger & Damasio, 1985).
Post-surgery, patient EVR, previously a successful businessman, was deeply impaired
at making even the simplest everyday decisions. EVR reported feeling trapped in a
cycle of deliberation over basic decisions, such as which pen to use to sign a contract,
or what cereal to select at the grocery store. Neuropsychological testing revealed that
EVR had no autonomic reactions to stimuli that typically generate strong affective
responses. That is to say, when a person is presented with a strongly affective image
they typically react with an increased heart rate and increased skin conductance
(Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Venables & Christie, 1973). In
contrast, when EVR was presented with affective images no visceral responses were
recorded, leading researchers to conclude that EVR was unable to generate feelings for
any visual images. EVR’s case supports our conjecture that the feelings or valence
generated during perception can bias choice behavior or decision-making. To reiterate,
if individuals fail to perceive the valence in objects, there is no basis for many of our
everyday, unconscious object-based decisions: for example, selecting the china cup
over the pottery mug, or the Mont Blanc fountain pen over the BIC ballpoint.
4
1.2 Object valence operates on a continuum
The most common framework for the majority of scientific experiments that
investigate affective perception has been to group objects into non-continuous,
qualitative categories. Objects are typically assigned one of three labels: positive (e.g.,
sports cars, roses, and cupcakes), negative (e.g., weapons, skulls, and grave stones),
or neutral (e.g., teacups, lampshades, and clocks) (Cuthbert et al., 2000; Lang et al.,
1998; Moriguchi et al., 2011; Rudrauf et al., 2008). This sort of qualitative organization
includes the assumption that objects assigned to the “neutral” category do not
generate a valence. We contend that this affective taxonomy is an oversimplification of
the complex variations in valence that form an integral part of our everyday perceptions
and interactions.
As an alternative we propose that object valence operates along a continuum
(Fig. 1A). Although this idea is not new (Colibazzi et al., 2010; Russell, 1980; Russell &
Carroll, 1999), other theorists have tended to focus on the extreme ends of the
continuum and objects with strong valence. In contrast, we focus on the fine grain
differences in micro-valence for objects close to the center. We posit that observers
perceive a valence for objects even in this region immediately surrounding neutral. That
is, the majority of nominally neutral objects such as chairs and clocks nonetheless
possess a micro-valence and so are always either slightly preferred or anti-preferred.
Indeed, when the scale at the center of the continuum is expanded we do not expect to
see large clusters of objects equivalent in their “non-valence”; instead, it is organized
much like the rest of the continuum, the only difference being the overall magnitude of
valence is weaker at the center than at the extreme ends (Fig. 1b – a plot in which the
order of objects in this expanded center portion of the continuum are based on data
5
collected in (Lebrecht & Tarr, 2010). Note that this model does not preclude the
presence of some objects falling at the center point of the continuum – such may give
rise to an absolutely neutral perception.
Figure 1.1 The valence continuum (1a) illustrates object valence ranging from strongly positive (shown in red) to strongly negative (shown in blue). In the center portion of the continuum objects generate a weaker perception of “micro-valence” that is either slightly positive or slightly negative. In Figure 1b the central portion of the continuum is expanded to reflect the same organization of Figure 1a, albeit with a lesser magnitude.
What is critical is that while micro-valence only represents a very small effect
size that accounts for a very small absolute region of the valence continuum, it is
integral in the perception of all objects, all of the time. When considered in this way,
micro-valence becomes an issue of exceptional magnitude. Undoubtedly,
understanding the processing of strongly affective objects is of great importance in
anchoring how the system works, but really how many times a week do you see a block
of gold or a blood stained weapon (barring the irreality of TV)? In real life, the vast
majority of our perceptual experience consists of mundane everyday objects.
In the study of affective processing, it is worth noting that although some of the
most common stimulus sets (e.g., the International Affective Picture System IAPS) have
!"#$%&'()*+,$+-,+.)/#(+0()*+,$+ 1&2#/#(+0()*+,$+
,+.)/#(+0"#$%&'()*+,$+ 1&2#/#(+0"#$%&'()*+,$+
3)4
354
6
been normalized along a continuum (Bradley & Lang, 1994; Lang, Bradley, & Cuthbert,
1997), researchers will often take these dimensional ratings and assign objects to non-
continuous categories (Lang et al., 1998) this reference is from the IAPS group using
their stimuli in a categorical experiment design). Grouping objects into non-continuous
categories for experimental reasons presupposes that objects in a particular group
possess equivalent affective properties (so a sports-car is equivalent to a cupcake if
they are both members of the positive group). This seems unlikely to be true for positive
or negative categories, but may particularly problematic for the neutral case where a
slight difference in micro-valence can actually lead to significant differences in behavior.
As one example, a person may own multiple different wine glasses. Yet there may be
one glass, perhaps with a slightly longer stem and deeper more rounded bowl, which
generates a more positive micro-valence, which could mean that that person selects
that particular wine glass with a greater frequency than the others. This suggests that
such subtle differences in micro-valence can actually lead to significant differences in
the frequency with which we interact with objects typically held to be neutral.
Supporting this reasoning, research in the affective processing of faces typically
relies on object properties organized along continuous dimensions (Todorov, Said,
Engell, & Oosterhof, 2008). For example, with respect to the affective properties of
faces, scientists have identified two key dimensions that explain, in a large part,
observers affective perception of faces: the dimension of trustworthiness (akin to
valence), whereby faces deemed to be more trustworthy are rated as more positive
than faces perceived to be less trustworthy; and the dimension of dominance (akin to
power), whereby faces perceived to be more dominant are rated as more negative than
faces perceived to be more submissive (Todorov, Baron, & Oosterhof, 1985). The key
point is that these are not categories (e.g., trustworthy or not); rather they are
7
intersecting continuous dimensions without a definitive neutral point. Moreover, the
invocation of these dimensions is part of an effort to account for the variance in
affective face perception. Similarly, our proposal for an object valence continuum aims
to account for subtle variations in object valence.
1.3 Understanding the origins of micro-valence
Understanding how everyday objects acquire a micro-valence forms a critical
component to our theory that a neutral valence rarely exists for most objects. We know
that during perception objects evoke a rich body of information (not necessarily
affective), including properties that are not an intrinsic part of the percept or the
percept’s category. For example, seeing an object can automatically activate an entire
network of information from memory (Bar, 2007). This network is derived from both
contextual experiences (Bar, Aminoff, & Schacter, 2008) as well as conceptual or
semantic knowledge (Haxby et al., 2001; Martin, Wiggs, Ungerleider, & Haxby, 1996;
Patterson, Nestor, & Rogers, 2007). That this concept applies widely can be seen in the
efforts by a group of historians to tell the history of the world in only 100 objects
(Schama, 2010): By selecting and describing key objects that elicit a wide range of
associations, historians were able to evoke critical periods in history. For example, one
particular object, the Roman Samian Bowl (Balmuildy Fort, Antonine Wall – 2nd century
AD), was bright red, engraved with human, animal, and floral figures. This bowl, made in
a workshop in France, was thought to have been used at banquets for Roman officers
and other high status individuals. The information attributed to this object told a story of
craftsmanship and Roman rule in France in 200 AD. In much the same we perceive a
richer network of information than that available in the percept, historians regard
objects as loci of archival information, acting as an access point for vast quantities of
8
knowledge and experience. Similarly, our perception of an object automatically
accesses a large network of information in our memories – valence is only one
component of an assortment of information, including semantic and conceptual
information computed during object perception.
Philosophers and cognitive scientists have been studying such associative
properties of objects for thousands of years (e.g., semantic memory, Aristotle, 1988;
Neely, 1977). Modern empirical investigation demonstrates that viewing a chair will
automatically activate related objects or concepts such as table, sofa, and, furniture
(Martin, 2007) and that automatic memory retrieval mechanisms serve to bring this
information rapidly online, without the perceiver directing any controlled memory
searches (Badre & Wagner, 2007). We know that object associations are generated, in
part, by seeing the object over and over again in certain configurations and contexts
(Bar, 2004; Bar, 2007). Therefore, we can assume that real-life experiences, which give
way to semantic associations, should not be thought of as exclusively cognitive.
As a common example, perceiving the table and chairs in your parents’ kitchen
undisputedly strengthens what are typically thought of as semantic associations: tables
go with chairs. However, the warm feelings one gets from sitting on a chair at the table
in your parents’ kitchen eating and drinking with family and friends are also encoded as
semantic associations (i.e., another kind of knowledge associated with a perceptual
object). Thus, we critically posit that valences are stored in the same network as other
semantic associations. Moreover, activation within this network during object
perception is a combination of all of these sources of knowledge, including factual,
contextual, and valence information (Barrett & Bar, 2009).
It is these associations, spontaneously reactivated during perception, that give
rise to an object’s micro-valence. To continue with the previous example, if my parents’
9
kitchen table was large and oak, whenever I saw a large oak table similar enough to my
parents’, it would automatically generate associations that would cause me to perceive
the other table as having a positive micro-valence. Note that the reactivation of such
associations may be subtle enough to never actually reach conscious awareness. The
principle that the contextual experience with an object informs the micro-valence is
evident in the micro-valence of telephones. Data from a recent study showed that
phones that resembled those used in the bedroom were rated as more positive than
those that were used in the office (Lebrecht & Tarr, 2010).
What makes this process more complex is that typically we have seen an object
in multiple contexts and with a variety of people, all of which will contribute to the
affective memories or associations that become automatically reactivated during object
perception. Micro-valence can thus be defined as the aggregate of this information
plus, as discussed below, any possible valence-relevant perceptual object properties.
This cumulative valence – accrued over our personal experiences – determines, in part,
whether a given object is ultimately preferred (positive micro-valence) or anti-preferred
(negative micro-valence).
Supporting this overall framework, there is strong evidence suggesting that
people rapidly attribute valence information to objects, which influences subsequent
perceptions (Bliss-Moreau, Barrett, & Wright, 2008; Ghuman & Bar, 2006; Murphy &
Zajonc, 1993; Zajonc & Markus, 1982). An individual does not even need to be aware of
the affective information paired with the object in order for it to influence perception at a
later point in time (Murphy & Zajonc, 1993). This was demonstrated in a study that was
able to change the valence of “neutral” Chinese characters (for non-native speakers) by
pairing them with a face displaying either a positive or negative facial expression. The
faces were presented subliminally so participants had no conscious memory of seeing
10
the face. Yet, the Chinese characters that were preceded by positive faces judged to be
more positive in following trials, and likewise for negative faces (Murphy & Zajonc,
1993). This simple demonstration illustrates that we are able to associate affective
information with objects, which, in the absence of awareness, can then bias perception
at a later point in time.
1.4 The perceptual components of micro-valence
Although one might think that micro-valences arise solely from learned
associations, there is evidence to suggest that low-level visual properties account in
part for an object’s perceived valence. So far we have discussed how learned attributes
contribute to an object’s micro-valence, but we also wish to address how the
perceptual features of objects may contribute to micro-valence. Can features such as
shape, curvature, color, and, symmetry lead to objects being perceived as positive or
negative in the absence of any affective associations? We can address this question by
looking at novel objects for which there are likely to be few already-established
associations.
Early work in this area shows that novel objects possess certain intrinsic
attributes, which lends the particular objects to be evaluated and named according to
particular perceptions. For example, when presented with a curved, squidgy object
participants are more likely to name it as “Bouba” when given the choice of Bouba or
Kiki. Alternatively, when participants see a jagged, spiky object they are more likely to
name it “Kiki” given the same choices (Kohler, 1929). These very early studies provide
evidence that we are able to evaluate and form judgments about novel shapes we have
never seen before based on very basic physical properties. A more recent study
confirmed these biases by demonstrating that when making fast “gut reactions”
11
participants preferred curved over sharp or jagged objects for both familiar and novel
objects (Bar, Neta, & Linz, 2006).
Although it is possible to make valence judgments on simple shapes (Bar et al.,
2006; McManus, 1980; Rentschler, Jüttner, Unzicker, & Landis, 1999), experimenters
report more reliable ratings for real world images as compared to abstract shapes
(Vessel & Rubin, 2010), from which we may conclude that experience based
associations contributes large amounts of information in the judgment of valence.
In sum, it seems most likely that micro-valences arise from an integration of
visual properties and learned associations. Moreover, these two attributes may
potentially interact in the following way – it might be easier to form positive associations
with objects already possessing “positive” perceptual features. For example, one might
more readily generate positive associations with a shiny, curved, symmetrical teapot,
whereas more readily generate negative associations with a dull, angular, asymmetric
teapot. Yet, this interaction between perceptual features and associations may also
work in reverse. There are data to suggest that color preference might actually arise
from the degree to which an individual likes an object with a particular color, so
participants would be more likely to prefer the color yellow to green if they prefer apples
to bananas (Palmer & Schloss, 2010).
1.5 How the perception of valence relates to the discourse of aesthetics
The study of object valence is closely related to the study of aesthetics. Yet,
whereas we are focused on the singular dimension of valence in perception, aesthetics
can be broadly defined as a discourse of how works of art are judged to be beautiful
(Baumgarten, 1750). Valence therefore can be considered to be one aspect of aesthetic
experience. Scientists interested in quantifying aesthetic experience have looked to the
12
low-level perceptual features in works of art in the hope of understanding what makes
them appealing (Ramachandran & Hirstein, 1999). As the discussion in the previous
section suggests, understanding the low-level features that give rise to a particular
percept may be informative not just for explaining aesthetic experience, but also for
understanding why we perceive an object to have a particular valence.
Aesthetics traditionally focuses on works of art in the museum that generate a
strong valence. However, the emerging field of the Everyday Aesthetics (Mandoki, 2007;
Saito, 2007) addresses how people make aesthetic judgments in their everyday
environment in much the same way as we propose that observers perceive valence in
common objects. What is interesting is how these seemingly banal everyday aesthetic
judgments actually have critical impacts on environmental and socio-political decisions.
For example, consider the desire to maintain lush, green lawns in dry, arid climates
simply because they generate a positive aesthetic experience. Often times, people will
go to great lengths using excessive amounts of water and chemicals to maintain the
appearance of bright green grass, regardless of the ecological consequences. The
same is true for drying clothes on a washing line. For some, washing lines have
developed a negative aesthetic and so there are actually neighborhood associations in
the United States that have banned families from hanging clothes to dry on outside
washing lines, in favor of using electric dryers, even on warm, breezy days.
1.6 Object valence and preference
Up to this point we have described the ways in which single objects acquire a
micro-valence during perception. But what happens when there multiple objects are
simultaneously present in a scene? As demonstrated by our initial example of reaching
for a mug from the cupboard, we often need to select a single object from multiple
13
competing objects. In other words, once an object’s valence has been computed
during perception, choice and decision-making systems must then compute the relative
micro-valences in order for us to make a selection (Constantino & Daw, 2010; Krajbich,
Armel, & Rangel, 2010; Litt, Plassmann, Shiv, & Rangel, 2011).
This decision process actually relies on more information than just the valence
of the object. The aforementioned decision-making systems must also account for the
predicted reward-value of each object given the context and the current goals of the
individual (Rangel et al., 2008). For example, a particular wine glass may have a highly
positive micro-valence, yet if a person is thirsty and their immediate goal is to reach for
a water glass, then they will select the glass based on function regardless of their
relative valences.
These object-based selections can be thought of as preference. Preference is
most commonly described at the level of behavior (Lichtenstein & Slovic, 2006). A
person that selects the cappuccino over the espresso is thought to have a preference
for cappuccinos. Yet, in order for someone to prefer a cappuccino over an espresso a
decision had to be made. As just explained, the relative information for each object
would need to be represented and computed. The micro-valence of an object may
provide one source of information that can help bias the decision one-way or another.
However, it is important to note that we hold that valence, as described here, is a
dimension of perception. That is, the integration of various sources of information
contributing to valence is a precursor to making these complex choices or decisions,
and not a consequence of the decision process itself. Thus, it can be said that valence
to contributes to our preferences rather than actually being our preferences.
1.7 Clinical implications for object valence
14
We have already seen from neuropsychological studies how damage to the OFC
can lead to devastating effects on simple choice behavior (Eslinger & Damasio, 1985).
However, when the connections between visual perception and affective processing
break down the effects can be extremely odd. Capgras delusion was first described
with respect to faces (see the review (Ellis & Lewis, 2001). In this disorder patients can
recognize individual faces without any difficulty, but they cannot associate any
emotional feelings to the face. The sensation of recognizing the face of a loved one, but
feeling nothing causes the patients to generate a delusion that helps make sense of the
situation: Patients report that their loved one has disappeared and been replaced by an
imposter (typically an alien or robot). This seems to be clearly a breakdown between
affect and vision, because: 1) these effects disappear if the patient speaks to his/her
relative on the telephone; 2) there are no reported deficits in either face processing or
emotional processing, but there does appear to be damage to the neural network
linking the two (Hirstein & Ramachandran, 1997).
Critical to our argument, it appears the Capgras disorder is actually a general
effect. For example, there are reported cases, similar to Capgras delusion, but for
inanimate objects that are referred to as delusions of inanimate doubles (Ellis et al.,
1996). In these cases, patients believe their possessions have been stolen and replaced
by replicas. One case study described patient, Mr. B, who reported his watch,
binoculars, wellington boots had been stolen and replaced by replicas (Anderson,
1988). The logic here is the same as in Capgras. When patients perceive a familiar
object and the expected valences and their consequent affective feelings are absent
they construct a story for why the object generates no feelings.
There also seem to be general disturbances with visual perception and affective
processing in other anxiety-related psychiatric disorders. As described throughout this
15
thesis, a large part of normal visual perception involves the automatic reactivation of
information associated with objects. In healthy individuals this information actually
assists with recognizing and interacting with particular objects (Bar et al., 2006; Barrett
& Bar, 2009). However, for some psychiatric patients the reactivation of this information
can be overwhelming and in some cases paralyzing.
In post traumatic stress disorder (PTSD), symptoms include vivid memories or
dreams of the trauma, recurrent thoughts and images, and emotional numbness
towards loved ones (DSM IV). Some of the basic mechanisms impaired in PTSD are
related to the mechanisms described with regard to object valence. For example, if a
person experiences a life-threatening accident with a nail gun, all of the affective
memories of that traumatic experience become associated with the nail gun. If that
person then perceives the nail gun at a later point in time the entire trauma associated
with the experiences reactivate. The intensity of the experience is magnified compared
to when affective associations are activated in response to objects during regular object
perception. In PTSD the associations become overwhelming, causing patients to re-
enter the fear-induced state of the initial trauma. Unfortunately, these symptoms can
easily transfer to other objects. For example, the patients described here would almost
certainly experience a traumatic response to the nail gun, but may also experience this
type of reaction to other related work tools, such as hammers and skill saws. PTSD
patients often adopt coping strategies to avoid these objects; in fact, criterion (B) of the
DSM IV for a diagnosis of PTSD includes “persistent avoidance of stimuli associated
with the trauma” (DSM IV). By understanding more deeply the complex behavioral and
neural processing surrounding object valence, we may be able to better understand
how affective object perception triggers recurrent trauma in PTSD and anxiety patients
more generally.
16
1.8 Why have the affective properties of everyday objects been underestimated?
It is surprising that everyday objects have been regarded as “neutral” for so
long. There appear to be three main reasons why researchers continue to overlook the
valence of seemingly neutral objects.
First, micro-valence is subjective. This is inherent in the very definition – that
affective associations, formed from memories and experiences, create an object’s
micro-valence. Clearly, people vary in their affective experiences with certain objects,
and the result is that a particular mug with a positive micro-valence for one person, may
have a negative micro-valence for another. The subjectivity of micro-valence leads to
high variance in the affective ratings of everyday objects, resulting in the sum or
average of micro-valence across a population converging to zero. This falsely suggests
that on average, objects are neutral when in fact most objects are never truly neutral for
any one person. This is evident from large image databases such as IAPS (Lang et al.,
1997), where the ratings for the neutral images in the set, such as fans, chairs,
newspapers and coffee mugs have a mean close to neutral, but a high standard
deviation, which suggests an inconsistency in these ratings across participants that
most likely captures the variation in micro-valence we are discussing.
Second, in psychological experiments objects with micro-valence are often
presented alongside strongly valenced objects. The positive micro-valence of a teapot
may be significantly underestimated if it has just followed the image of an ice-cold
bubbling glass of champagne or a large bundle of $100 bills. In terms of the valence
continuum (Fig. 1a), the presence of strong objects in an experiment actually pushes
objects with a micro-valence more towards the central region of the continuum,
resulting in people perceiving them as being more neutral. In order to not underestimate
17
the valence of supposedly “neutral” objects, such items need to be presented prior to
the condition(s) containing strongly valenced objects.
Third, because the evaluation of micro-valence is automatic and unconscious
observers may be unaware that they are making choices or decisions based, in part,
from the valence of objects. As discussed throughout the paper, the perception of
micro-valence arises in part from the affective associations that are spontaneously
reactivated during perception. These affective associations or feelings may never
actually reach conscious awareness, yet research has shown this is not necessary for it
to influence behavior (Litt et al., 2011; Murphy & Zajonc, 1993). When asking a
participant explicitly which mug, or clock, or lampshade has the most positive valence
they may not have conscious access to this information, and so may report that all of
the objects are neutral in valence.
1.9 Discussion
The functional significance of objects with strong valence is highly intuitive; we
dislike objects that indicate danger or threat, such as angry tigers or moldy food, and
we like objects that indicate sustenance or pleasure, such as appetizing food or
attractive mates. A lamp or a teacup is neither threatening nor life promoting, so why
would we have evolved a mechanism to associate subtle or micro-valences with
everyday objects?
These subtle micro-valences function to optimize our ability to either select or
orientate towards objects with a positive micro-valence and away from those with a
negative. Throughout the day we make multiple “thoughtless” selections: what mug to
use for our morning coffee, what pen to sign with, and what bottle of water to purchase.
We propose that these decisions are facilitated by the micro-valences computed during
18
perception and that we then use these micro-valences in situations of uncertainty to
automatically select or orientate towards those objects with the most positive micro-
valence, and away from those objects with the most negative valence.
Of course, these sorts of issues have not been lost on the product design and
marketing communities. For example, Donald Norman, an expert in the psychology of
product design, has argued that the affective properties, or as we would say, the micro-
valence, enhance the usability of particular objects (Norman, 2003). Illustrating this
point, ATMs designed to have more positive valence have been reported, cross
culturally, to be easier to use, which suggests that valence can affect not only object
selection, but also object function (Hekkert, 2006).
In summary, we contend that our perception of the world is never “neutral”.
Ultimately, we are social creatures that, through a variety of contextual experiences,
combined with emotions, create a visual world animated with affect. As observers we
must be able to decode the multitude of perceptual, affective, and semantic information
present in our visual field. One way to solve the affect part of this equation is to
automatically evaluate the valence of the majority of objects present in the scene during
perception. We suggest that, as perceivers, much in the same way that we cannot help
but see the shape, size, or color of objects, we cannot help but see the valence in
objects. That is, valence is not a label applied after the fact to perceptual entities, rather
it is an integral element of perception with the same mental status as any other object
attribute.
19
2. CHAPTER TWO
The Affective Lexical Priming Score
Introduction
A long history of research in affective priming has demonstrated that we extract
valence, rapidly and automatically, from the world around us (Bargh, Chaiken,
Govender, & Pratto, 1992; Fazio, Jackson, Dunton, & Williams, 1995; Fazio,
Sanbonmatsu, Powell, & Kardes, 1986; Hermans, De Houwer, & Eelen, 1994; for
reviews see Fazio, 2001; Klauer & Musch, 2003; Klauer, 1997). Affective priming – faster
or more accurate processing when one affective item is preceded by another affective
item with the same valence direction – has been demonstrated with a variety of different
stimulus types (e.g. words (Fazio et al., 1986); scenes (Avero & Calvo, 2006; Calvo &
Avero, 2008); faces (Lebrecht, Pierce, Tarr, & Tanaka, 2009); odors (Hermans, Baeyens,
& Eelen, 1998). Remarkably, affective priming occurs even when the prime and target
are of different forms (e.g. face prime, word target) (Fazio et al., 1995; Lebrecht et al.,
2009; Vanderwart, 1984); indicating, that valence is not tied to a particular domain, but
rather is represented in a domain-general manner. In this paper, we present a cross-
modal affective priming paradigm known as The Affective Lexical Priming Score (ALPS)
that uses the novel combination – for affective priming – of a visual prime paired with a
lexical decision. Given that affective priming is such a well-established, well-researched
area, our goals are not to again demonstrate the presence affective priming; instead,
we aim to understand how automatic affect evaluation – the mechanism underlying this
type of priming – contributes to socio-cognitive and affective processing. To effectively
address this question, it is often necessary to use tasks that can be administered
alongside other experiments in a single session. With this in mind, we designed ALPS
20
to be a simple, fast-to-administer paradigm suitable for use in conjunction with other
tasks. In an effort to make ALPS as robust as possible, we developed a controlled set
of object primes that were normed for valence (Experiment 2.1), and a controlled set of
word and non-word targets (Experiment 2.2) that were normed for response times. We
then employed these stimuli in the ALPS paradigm and report our findings in
Experiment 2.3.
Priming effects were first demonstrated in psycholinguistic experiments
examining lexical processing and semantic memory retrieval (Meyer & Schvaneveldt,
1971; Neely, 1977). In semantic priming, the presentation of a stimulus “primes” or
facilitates the speed of response to a semantically related target word because the
semantic associations of the prime and target are overlapping. For example, when
participants are presented with the word doctor they are faster to identify to the word
nurse as a valid word, compared to, for example, the word tree, because the words
doctor and nurse share overlapping semantic associations (Meyer & Schvaneveldt,
1971). This paradigm was first adapted for studying affective priming by (Fazio et al.,
1986). Here the logic is the same, yet instead of overlapping semantic associations
leading to priming, priming effects arise from overlapping affective associations. For
example, participants would be faster to identify the word ice cream having been
primed with the word gold, because both the prime and target words possess
congruent or matching valences. Conversely, participants would be slower to identify
the same word – ice cream – having just been primed with the word gravestone,
because the two words have incongruent or mismatched valences (Fazio et al., 1986).
These and other findings suggest that valence, independent of semantics, can prime an
individual’s response to an otherwise unrelated word.
21
The actual task used by (Fazio et al., 1986) in their initial affective priming
experiment was modified from lexical decision (as used in (Meyer & Schvaneveldt,
1971) to a word evaluation judgment, where participants categorize a word as positive
or negative. For the most part, researchers investigating affective priming have
continued to use word evaluation judgments to assess the automatic recruitment of
valence (Aguado, García-Gutierrez, Castañeda, & Saugar, 2007; Fazio et al., 1995;
Hermans et al., 1994) (but see Hill & Kemp-Wheeler, 1989). Despite a large number of
affective priming studies conducted over the past two decades (for a review, see Klauer
& Musch, 2003), none have employed the experimental structure of ALPS that
combines image primes with a lexical decision. In a summary table of more than 80
affective priming studies, (Klauer & Musch, 2003) did not report a single study that used
the combination of an image prime with a lexical decision. There are, however, studies
that employ lexical decisions in affective priming, but only in paradigms where words
serve as both the prime and the target (Hill & Kemp-Wheeler, 1989).
If affective priming demonstrates the extraction of valence from a given
stimulus, then the parameters under which this priming occurs help explicate the
mechanisms underlying this phenomenon. The main parameter governing affective
priming is the stimulus onset asynchrony (SOA), which is the time between the onset of
the stimulus and the onset of the target. A number of previous studies have shown that
affective priming effects can be observed at SOAs as short as 300ms, but no longer
than roughly 800ms (Aguado et al., 2007) to 1000ms (Fazio et al., 1986; Hermans, De
Houwer, & Eelen, 2001; Hermans et al., 1994; De Houwer, Hermans, & Eelen, 1998).
Since 300ms is a relatively brief time with which to consciously process information
(Posner & Snyder, 1975), these findings indicate that affective priming relies on
automatic processing, hence affective priming is often referred to as Automatic Attitude
22
Activation (Fazio, 2001). Other measures that also rely on automatic processing, such
as the Implicit Association Test (IAT) (Greenwald, McGhee, & Schwartz, 1998; Nosek,
Banaji, & Greenwald, 2002), provide further evidence that individuals rapidly generate
affective associations implicitly and without conscious control and, in doing so, form an
implicit affective judgment about a particular stimulus. Taken together, the evidence
from both SOA manipulations and from tasks such as the IAT suggests that the
mechanisms in place for extracting valence from stimuli rely on automatic processing
networks that retrieve or reactivate affective associations associated with the stimulus.
As already discussed, affective priming can be seen for a variety of different
stimulus types, including faces (Fazio et al., 1995), scenes (Avero & Calvo, 2006; Calvo
& Avero, 2008), color images of objects and animals (Hermans et al., 1994), black-and-
white line drawings (Giner-Sorolla, García, & Bargh, 1999), words (using evaluative
decision tasks (Fazio et al., 1986) or lexical decisions (Hill & Kemp-Wheeler, 1989), and
even odors (Hermans et al., 1998). From these studies and others, we can say with
confidence that individuals have the ability to rapidly extract valence from a variety of
different stimulus domains. In particular, in ALPS we use visual object primes because
we are interested in how participants extract valence in the visual domain, as well as
how such valence information interacts with other knowledge to form judgments, make
decisions, and plan behaviors.
It is worth nothing that human faces are a strongly affective visual cue, which
makes them a reliable stimulus class for affective priming studies; in particular, some
studies focus on the emotional expression of the face as a valence cue (Aguado et al.,
2007), while others focus on the race (Fazio et al., 1995; Greenwald et al., 1998;
Lebrecht et al., 2009). Affective priming can also occur with visual stimuli other than
faces, as indicated by studies that report affective priming for more complex visual
23
scene images (Avero & Calvo, 2006; Calvo & Avero, 2008). Avero et al. (2006) used an
image-based priming paradigm in which participants were presented with an image of a
scene and then required to categorize the valence of a subsequently presented target
scene. The authors controlled for the perceptual and semantic similarity across the
image pairs by pairing scenes containing animals with scenes containing people.
Controlling for the perceptual similarity of the prime and target it was found that highly
dissimilar images still produced high levels of affective priming. Thus, valence appears
to be a metric independent of low-level semantic and perceptual properties.
Surprisingly, observers can automatically extract the valence from the visual
environment without directed attention (Aguado et al., 2007; Calvo & Avero, 2008;
Calvo & Lang, 2004). Aguado et al. (Aguado et al., 2007) report affective priming for
emotional faces matched in valence when participants make a gender – not an
emotional – judgment on the face. This illustrates that even when attending to a neutral
feature of the face (gender) participants cannot help but automatically compute the
valence associated with a particular expression. Calvo et al. (Calvo & Lang, 2004) found
that observers can even extract the valence from an image when they are not looking
directly at it. Participants were instructed to fixate on a central cross while images were
presented in the periphery, some with a strong valence and others with a relatively
neutral valence. Results showed that participants were significantly more likely to make
their first eye movement to an image with a strong valence compared to a relatively
neutral image. From this finding, Calvo et al. 2004 hypothesized that participants were
extracting the valence of the peripheral images using covert attentional strategies while
overtly attending to the central fixation cross. This was confirmed in a second study
that reported affective priming in peripheral vision while participants engaged in an
unrelated task at the fovea (Calvo & Avero, 2008). The prime and target images were
24
presented 2.5 degrees away from central fixation and while participants were engaged
in a task at the fovea, they were asked to simultaneously report the valence of the
target image presented in the periphery. Participants were faster to respond when the
valences of the prime and target images matched. This finding illustrates that an
observer’s ability to extract valence from perceptual input can occur automatically and
in the absence of direct attention.
The majority of affective priming work that uses visual primes that are not faces
relies on the International Affective Picture System (IAPS) (Lang et al., 1997). IAPS is a
large image database used extensively in the field of affective and emotional
processing. Most IAPS images depict highly complex scenes, which include a multitude
of faces, animals, and objects. Given the variety and complexity of information present
in most IAPS images, it remains unclear how participants rapidly compute a single
unified valence that is nominally detected during basic priming paradigms.
To avoid this problem, we investigated affective priming for the valences
associated with single, non-living objects. Our particular affective priming paradigm,
known as the Affective Lexical Priming Score (ALPS), uses these visual objects as the
primes for a lexical decision task. This new set of affective object primes were normed
by an independent group of participants to be either strongly positive or strongly
negative, the methods for which we describe in Experiment 2.1. Likewise, the target
words used in ALPS were normed, the results of which are described in Experiment 2.2.
We then report findings for a specific instantiation of ALPS in Experiment 2.3.
2.1 Object Norming
When presenting observers objects of a particular valence, as experimenters,
we must be confident that the average participant is going to perceive the object to
25
have the valence that we expect. We address this issue by normalizing the stimulus set
with an independent group of participants in Experiment 2.1.
It has long been suggested that there are two systems for processing
information: a fast, associative, automatic system, and a slow, controlled, deliberative
system (Chaiken & Trope, 1999). Given that affective priming relies on the former, our
goal here was to obtain valence ratings for object primes using a task that was rapid
and automatic. That is, briefly-presented stimuli seem most likely to evoke a consistent
valence.
As discussed earlier, we chose to employ single non-living objects that did not
require participants to engage in in-depth semantic processing (in contrast to the more
complex IAPS images). As such, we assembled a large set of strongly positive and
negative objects. On each trial, a single object was displayed above a black line.
Participants were required to click a point on the line to indicate how pleasant or
unpleasant they perceived an object to be. We anticipated that by (a) presenting
objects rapidly, (b) restricting the response time window, and (c) not requiring
participants to generate a verbal report, that participants would be most likely to
provide an automatic response based on their “gut feeling” of valence. At the same
time, we acknowledge this task is not an entirely automatic in that participants are
explicitly rating the valence of the objects. However, it is arguably more automatic than
a numeric or verbal rating provided during an unlimited response window.
Participants:
Participants were fifteen individuals (seven female/eight male) from the Brown
University community (mean age = 25.46 years; s.d. = 3.5 years). The protocol was
26
approved by the Institutional Review Board at Brown University. Participants gave
informed written consent and were compensated for their time by monetary payment.
Stimuli:
263 objects were presented in the experiment: based on visual inspection we
deemed 140 of the objects to have a positive valence and a 123 of the objects to have
a negative valence. All objects were colored images of non-living objects that were
centered in 400 x 400 pixel region presented in the center of a white screen.
Thumbnails of selected images are included in the Appendix. Stimuli were presented
using Matlab (Mathworks Inc., Natick, MA, USA) and Psychtoolbox (Kleiner, Brainard, &
Pelli, 2007; Pelli, 1997) on an LG LCD monitor and viewed (1024x768 resolution) at a
distance of 60 cm.
Task:
On each trial, a single object was presented in the center of the screen above a
single black line. Participants were instructed to use the cursor to click a point on the
line to indicate how pleasant or unpleasant they perceived the object to be. They were
told that the far left of the line indicated “most negative” and the far right of the line
indicated “most positive”. During the 45 practice trials we presented symbols at either
end of the line to remind participants of the direction of pleasantness, that is, (–) was
presented at the far left and (+) was presented at the far right. These symbols were
removed during the actual experiment so as not to direct the participants’ attention
towards the extremes of the line.
Each object was presented for a maximum duration of 3 seconds. Participants
were free to respond at any time during the 3 seconds, once they responded the next
27
trial began with the cursor automatically returning to the center of the line. The time limit
in this task encouraged participants to make their evaluation as rapidly and
automatically as possible.
Results:
Objects were assigned a continuous numerical value that corresponded to a
point on the line, so objects that were rated at the furthest left point were assigned a 0
and objects rated at the furthest right point were assigned a 1. We took an average of
the ratings across participants for each object and object’s with a mean rating of 0.6 or
higher were categorized as positive, and object’s with an average rating of 0.4 or lower
were categorized as negative. Objects that did not satisfy these cutoff boundaries were
removed from the stimulus set. Of the objects in the initial set, 120 of the most positive
and 120 of the most negative objects were selected to be included in the stimulus set
employed in ALPS. Examples of the 240 objects used in ALPS are presented in the
Appendix alongside their mean rating and standard deviation.
2.2 Word Norming
In picture-word affective priming paradigms, participants are typically required
to either pronounce the word aloud (Giner-Sorolla et al., 1999), or make an evaluative
decision about the word’s valence (Fazio et al., 1995) (i.e. decide whether the word is
positive or negative). In ALPS we use a novel combination of a visual prime and a
lexical decision task. One major limitation of all tasks that require some processing of
affective words is that participants are faster to process positive words compared to
negative or neutral words (Stenberg, Wiking, & Dahl, 1998; Unkelbach, Fiedler, Bayer,
Stegmüller, & Danner, 2008). This is certainly true for lexical decision tasks, where
28
participants are on average faster to make lexical decisions for positive words than for
negative or neutral words (Ferraro, Christopherson, & Douglas, 2006; Hill & Kemp-
Wheeler, 1989; Unkelbach et al., 2008; Roos et al., under review). This main effect of
word valence makes (Prime x Target) interactions typically reported in affective priming
studies harder to interpret, and sometimes less significant. However, these effects are
not always immediately obvious in the literature in that experimenters often report
priming effects by contrasting average response latencies for affectively congruent
versus incongruent prime-target pairs. Although usually done “for the sake of simplicity
and clarity,” (Wentura, 2000) reporting results in this manner obscures any effects due
to target valence alone. Dijksterhuis and Aarts (2003) reanalyzed data from five previous
affective priming studies using a variety of tasks in which response times for all possible
combinations of prime and target were reported. They found that in all five studies,
participants responded faster to positive than to negative targets (independent of prime
valence). This finding was confirmed by Unkelback et al. (2008), who reanalyzed data
from seven previous studies and found the same pattern of results: participants
responded faster to positive than to negative words.
The difference in response times for positive and negative words in these
priming tasks is a significant issue for affective priming. Experimenters have attempted
to control for these differences using standard lexical control measures (i.e. word
frequency and word length), although these have been mostly unsuccessful because
they do not address the specific issues surrounding word valence (Unkelbach et al.,
2008; Unkelbach et al., 2010). Given that lexical controls do not appear to eliminate this
effect, we decided to take a somewhat novel approach to controlling for this factor in
our word sets. More specifically, because ALPS relies on a lexical decision task, and we
anticipate observing a main effect of positive words, we chose to select words based
29
not on frequency and length, but based on how quickly (on average) participants can
make a lexical decision for that word.
To this end, we employed a lexical decision task to measure participants’
response times for affective words in the absence of any visual prime. The goal of this
task was to identify a subset of words that shared the same mean and distribution of
reaction times across valence in an effort to eliminate the typically-observed response
times advantage for positive words.
Participants:
Participants were nineteen individuals (ten female/nine male) from the Carnegie
Mellon University community (mean age = 23.95 years; s.d. = 4.62 years). The protocol
was approved by the Institutional Review Board at Carnegie Mellon University.
Participants gave informed written consent and were compensated for the time either
by monetary payment or course credit.
Stimuli
There were 1200 letter-strings used in this experiment, 600 real words and 600
non-words. Of the real words, 200 were positive in valence, 200 were neutral, and 200
were negative. All of the real words were selected from the Affective Norms for English
Words (ANEW) database (Bradley & Lang, 1999), which contains words that have been
normed for valence on a 9-point scale from most unpleasant (1) to most pleasant (9). In
the current study, positive words had valence scores ranging from 7.12 to 8.72, neutral
words ranged from 4.98 to 6.32, and negative words ranged from 1.25 to 2.93.
Non-words were randomly generated using the ARC Non-Word Database
(http://www.maccs.mq.edu.au/~nwdb/). The non-words were deemed: (a)
30
pronounceable, and (b) phonetically distinct from any pre-existing word in the English
language.
All words were presented in black Helvetica font, 30 pt. Words were
presented in the center of a white screen using Matlab (Mathworks Inc., Natick, MA,
USA) and Psychtoolbox (Kleiner et al., 2007; Pelli, 1997) on an LG LCD monitor and
viewed (1024x768 resolution) at a distance of 60 cm.
Task:
Participants were instructed to decide whether they thought the letter-string
presented was a real word or a non-word. The letter-string appeared for a maximum
duration of 1400ms. Participants could respond either while the letter-string was on the
screen or during a 200ms response window following its erasure. Following a response,
a red fixation cross appeared at the center of the screen in preparation for the
upcoming trial. After every 100 trials, there was a self-paced break to ensure that
participants remained focused for the duration of the experiment. Trial order was
randomized across participants.
Results
From the original set of 600 words, 6 words (5 negative, 1 neutral) were excluded
from analyses because more than 50% of participants responded incorrectly. An
additional 24 words (8 positive, 6 negative, 10 neutral) were excluded because the
average reaction times were two standard deviations above or below the group mean.
The following results report on the 570 remaining words.
Mean reaction times for each condition are presented in Table 2.1. A one-way
analysis of variance (ANOVA) revealed a significant difference between conditions (F (2,
31
567) = 13.40, p < .001). Post-hoc (Least Significant Difference) comparisons revealed
that participants were significantly faster to identify positive words than they were to
identify both neutral (mean difference = -8.66, p = .014) and negative words (mean
difference = -18.12, p = .000), and that participants were significantly faster to identify
neutral words than negative words (mean difference = -9.47, p = .007). These results
indicate that participants are fastest to identify positive words, slowest for negative and
neutral words fall somewhere in between.
Valence N M (RT) SD
Positive 200 559.97 30.98
Negative 195 578.09 36.10
Neutral 199 568.85 34.92
Table 2.1 Mean (M) Reaction Times (RT) and Standard Deviations (SD) for Positive, Negative, and Neutral Words.
Assembling a Word Set
Given the reported differences in mean response times for the three word
categories, we selected a set of words for each condition that contained only words
with roughly equal response times. To do this we selected words that had a mean
response time within 1 second of the reaction time for the words in each of the other
categories. Words with response times that did not fit this criterion were discarded. This
limitation left us with 92 words per category. Because we only required 70 words per
condition for ALPS we manually selected 70 of the words with the strongest valence for
the final word set. Response times for this new word set were submitted to a second
analysis: mean reaction times and standard deviations are presented in Table 2.2 and
32
an ANOVA confirmed that there were no significant differences between conditions (F
(2,207) = 0.46, n.s). In addition, the distribution in response times for the individual
words in each condition were found to be overlapping.
Finally, the non-words selected for use in Experiment 2 were matched for average
word length with the real words. An ANOVA confirmed that there were no significant
differences between the lengths of words and non-words (F (1, 418) = .08, p = .77).
Valence N M SD
Positive 192 558.94 28.77
Negative 189 563.18 28.31
Neutral 189 562.71 28.43
Table 2.2 Mean (M) Reaction Times (RT) and Standard Deviations (SD) for Positive, Negative, and Neutral Words in Final Word Set.
2.3 The Affective Lexical Priming Score
ALPS is an affective priming procedure that combines an image prime with a
lexical decision task. We have used ALPS successfully in previous studies with face
primes (Lebrecht et al., 2009), and here we examine whether the effects are present
when you substitute non-living objects for faces, the results of which would suggest
ALPS is measuring a more domain-general process of valence evaluation for visual
input.
We argue that there are clear advantages to a measure of implicit attitudes that
pairs a visual prime with a lexical decision. First, participants are naïve to the true goals
of the experiment because the nature of the task contains no direct reference to
valence, unlike other tasks that require participants to make an explicit word evaluation
33
judgment, which unavoidably draws their attention to the valence of the target word
(Fazio et al., 1995; Fazio et al., 1986; Greenwald et al., 1998). In ALPS, because the
lexical decision task does not draw participants’ attention to the goals of the study
there is no need for a series of cover experiments to divert their attention, as is
common for experiments using the word evaluation task (Fazio et al., 1995). Thus, a
second advantage is the absence of any cover experiment, which makes ALPS a short,
easy-to-administer task suitable for use in conjunction with other experiments. As such,
ALPS may be suitable for studying how the evaluation of valence relates to other
cognitive processes, which may also need to be assessed in the same experimental
session. Third, as a cross-modal priming paradigm (and unlike image-image priming
paradigms), there is no chance in ALPS that the perceptual similarity of the prime and
target can account for the priming effects reported. This is not to say experimenters
cannot report reliable effects in image-image priming paradigms, but rather that
experimenters must go to great lengths to control for potentially confounding
similarities between the prime and target image pairs (Avero & Calvo, 2006).
Participants
Participants were twenty-two individuals (eleven female/eleven male) from the
Carnegie Mellon University community (mean age = 21.86 years; s.d. = 1.58 years). All
participants were native English speakers. The protocol was approved by the
Institutional Review Board at Carnegie Mellon University. Participants gave informed
written consent and were compensated for the time by monetary payment.
Stimuli
34
Primes were 420 unique color images (400 x 400 pixels) of non-living objects.
The object primes used in word trials were derived from Experiment 1a. The objects
used in the non-word trials were not normed, but were identical in size, style, and
valence to objects in word trials. Because these objects served as primes only in non-
word trials (which would ultimately be discarded), it did not pose a significant problem
that they had not been normed for valence. Moreover, since there are no noticeable
differences between the object sets, it is unlikely that a naïve participant would be able
to predict the nature of the upcoming word type based on the appearance of the prime.
Target words and non-words were derived from Experiment 1b. There were 210
words and 210 non-words. Of the 210 words, 70 were positive, 70 neutral, and 70
negative (for details on how word sets were controlled see Experiment 1b).
Task
Participants were instructed to attend closely to the object prime and then make
a lexical decision on the subsequently presented target letter-string. Participants were
motivated to attend to the prime because they were told that it was their cue that the
letter-string was about to appear. The object prime was presented in the center of the
screen for 250ms, followed by a 50ms Inter-Stimulus Interval (ISI), after which a target
letter-string appeared for 500ms (for details see Figure 1). Participants were able to
respond while the letter-string was on the screen or during a 1 second response
window that followed. Participants pressed the “1” key for word responses and the “2”
key for non-word responses; this button assignment was counter-balanced across
participants. The precise pairing of object primes and target letter-strings was
randomized across participants.
35
Figure. 2.1 The schematic of ALPS depicted in Figure 2.1 presents image primes that are either positive or negative non-living objects for 200ms, followed by a 50 ms fixation cross. After this 250ms signal onset asynchrony (SOA) a target letter-string appears, which is either a positive, negative, neutral, or non-word, and participants are required to make a lexical decision: pressing 1 on the keyboard for “word” and 2 for “non-word”. The keyboard responses are counter-balanced across participants. Participants are free to respond during the presentation of the letter-string (500ms), or during a fixed response window following (1000ms). The following trial then begins with the presentation of the object prime.
Results and Discussion
One participant was excluded from analysis because more than 25% of her trials
were answered incorrectly. A second participant was excluded because their average
response times were more than two standard deviations above the group mean. For the
remaining twenty participants, incorrect responses were removed and response times
two standard deviations above or below the mean were omitted from further analyses.
On average, participants responded correctly for 91% of non-word trials and 94% of
word trials.
Results were consistent with commonly-observed affective priming effects. As
shown in Figure 2.2, participants were faster to identify a word when the object and the
word matched in valence (Congruent) as compared to when the object and the word
were did not match (Incongruent).
!"#$
%
&'"&
%
()*+
%
,"-+
%
36
Figure 2.2 The facilitation scores reported in Figure 2.2 confirm that participants are significantly faster to respond during congruent trials (a match in prime and target valence) compared with incongruent trials (a mismatch in prime and target valence). Response time facilitation, measured in milliseconds, is represented on the vertical axis. Here congruent trials show a speeded response or facilitation relative to a neutral word baseline (i.e. a score above zero), whereas, incongruent trials show a slowed response or inhibition relative to a neutral baseline (i.e. a score below zero). Error bars represent the standard error of the mean.
All statistical tests presented were based on response times normalized for
group-level variations. We did this by subtracting an individual’s response time for
neutral word trials from Congruent or Incongruent trials. For example, positive
prime/neutral word trials were subtracted from positive prime/positive word trials to
measure the degree of facilitation; positive prime/neutral word trials were subtracted
from positive prime/negative word trials to measure the degree of inhibition. The same
subtractions were done for negative prime trials. The reason that neutral words were
presented with either positive or negative object primes (and subtracted accordingly),
rather than with neutral image primes, is because we wished to obtain the most
accurate response time baseline for both positive and negative prime trials separately.
A matched pairs t-test on these scores showed that response times were
significantly faster for Congruent prime-target pairings than Incongruent pairings (t(39) =
Congruent Incongruent
-10
-5
0
5
10
15
20
Trial Type
Facili
tation (
ms)
37
2.06, p = .04) (for details see Figure 2.3), which is supported by a repeated measures
analysis of variance (ANOVA) that revealed a marginally significant interaction of prime
valence and target valence (F (1,19) = 4.18, p = .055), illustrating that participants were
faster to respond when the prime and target matched in valence (Figure 2.3).
Figure 2.3 The crossover interaction in Figure 2.3 demonstrates that participants are faster to make lexical decisions for words where the prime and target match in valence, compared to when they are mismatched. In this graph, target word valence is presented on the horizontal axis and prime valence is indicated by either a dashed line for positive object primes or a solid line for negative object primes. The facilitation in response time is measured in milliseconds and represented on the vertical axis; it is computed by subtracting response times from a neutral word baseline. Scores above zero indicate a facilitated response, whereas scores below zero indicate an inhibited response.
There was no main effect of prime valence or target valence, indicating that we
were successful in our efforts to significantly reduce the main effect of positive words.
Yet despite the main effect being non-significant it should be noted that there was still a
trend towards significance (F (1, 19) = 3.17, p = .09). The effects of this can be seen in
Figure 2.3, where it appears the interaction is primarily driven by positive-positive
priming with very little negative-negative priming. However, we posit that positive-
Positive Negative-15
-10
-5
0
5
10
15
20
25Positive Prime
Negative Prime
Target Valence
Facili
tation (
ms)
38
positive priming is slightly enhanced by the speeded response to positive words,
whereas the negative-negative priming is slightly inhibited by the slower response
times. As described in Experiment 2.2 we only selected words that were matched in
average response times for a lexical decision in the absence of an affective prime. In
Experiment 2.3, this non-significant trend hints at an interaction between the overall
processing speed advantage for positive words and affective primes.
General Discussion
Our goal here is to present ALPS, a novel cross-modal affective priming
paradigm that combines non-living, single object primes with a lexical decision task. In
Experiments 2.1 and 2.2, we outlined our procedures to normalize the image and word
sets across an independent group of participants. With this highly controlled stimulus
set, in Experiment 2.3 we report results for ALPS that are consistent with the most
commonly observed findings of other measures of affective priming: participants
responding faster to positive words following a positive image prime and faster to
negative words following a negative image prime. This finding is consistent with an
extensive literature on affective priming demonstrating that observers automatically
extract valence from a variety of different stimulus types (Avero & Calvo, 2006; Fazio et
al., 1995; Hermans et al., 1998). Yet our finding is notable in that our task makes no
reference to valence whatsoever, unlike the more common word evaluation tasks, and
that the effects we report occur across different stimulus domains (visual objects and
lexical tokens).
Of particular note, in creating ALPS we managed to significantly reduce the
main effect of word type, but we did not manage to eliminate it entirely. That is,
participants’ speeded responses to positive words over and above those for negative
39
words appears to be highly robust. Such findings are in line with a literature that shows
participants are faster to detect stimuli with negative valence (particularly threatening
stimuli (Öhman, Lundqvist, & Esteves, 2001), but are in fast faster to process stimuli
with positive valence (Unkelbach et al., 2008), for example, tasks that involve
recognition or categorization. These differences in affect occur for words (Stenberg et
al., 1998), faces, and even schematic faces (Leppänen & Hietanen, 2004), suggesting it
is a property of valence and not low-level visual features. To account for these
differences, Unkelbach et al. (2008) proposed the Density Hypothesis. According to this
theory, positive information is more similar and therefore more densely clustered in
memory. Unkelbach et al. propose that we are more familiar with items of positive
valence, which leads to more associations and also a denser representation in the
lexicon. For example, multidimensional scaling shows that positive words, such as
“pretty” and “kind” are often closer in the lexicon than negative words like “ugly” and
“unkind”. This organization means that it takes participants less time to identify whether
a positive word is in fact a real word, because the positive words are “closer” to one
another and therefore reinforce one another when activated, which accounts for why
participants are faster to process positive words.
In light of our findings using ALPS, we postulate: a) individuals can extract
valence from common visual objects (e.g., not faces or extremely affective IAPS
scenes); b) they can do so automatically, without any intention or directed strategy; and
c) this valence metric, once generated, can influence processing times for accessing
one’s lexicon, a cognitive system typically assumed to be independent of affective
processing. The finding that the rapid and automatic perception of an object’s valence
can impact processing times in a lexical decision task suggests that, as a field, we
should reassess how valence interacts with other cognitive systems. Given our present
40
results, we suggest that within 300ms participants have perceived the valence of a
visual object and that this information has been transmitted to the lexicon such that it
can influence a task nominally unrelated to valence. Thus, we posit that valence is not
tied to a particular stimulus domain, but rather manifests across domains. Taken
together with cross-modal studies of affective valence (Giner-Sorolla et al., 1999;
Hermans et al., 1998; Sollberge, Rebe, & Eckstein, 2003), our present results speak to
the extent to which cognitive and affective systems, once thought to be separate, are in
fact highly interconnected.
Why might this be the case? That is, why should it be so that the valence of
objects routinely influences our cognitive and perceptual states? Consider that as
demonstrated here, as well as in other affective priming tasks, participants perceive the
valence of a stimulus and that this valence apparently prepares participants for an
upcoming response, that is, a lexical decision in the case of ALPS. This leads us to a
more specific question – when we perceive valence in objects in the real world, what
does this information prepare us for? Monitoring our environment and predicting likely
events can be highly adaptive, which makes the ability to rapidly detect threatening or
advantageous stimuli a desirable trait. Supporting this framework, recent evidence
suggests that prediction plays a key role in visual object recognition (Bar, 2007), where
the predictive signals can take the form of either visual input or affective responses
(Barrett & Bar, 2009). Thus, the perception of valence could contribute to approach-
avoid behaviors, to judgment formation and stereotypes, or more generally, to
expectations about upcoming events. Thus, as odd as it might seem that the valence of
a visual object can influence how rapidly observers can judge a letterstring to be a
word, the large advantage conferred by the automatic evaluation of valence renders the
automaticity and the ubiquity of valence across domains less mysterious.
41
In sum, we have developed ALPS with the intention of better understanding how
affective perception prepares participants for subsequent cognitive processing. As
such, we had as one goal, the need to study affective priming alongside other within-
subject measures, often times in a single experimental session. In order to use an
affective priming measure in a multistage experimental design one’s assessment tool
needs to be both fast to administer and stand alone with respect to cover stories and
cover experiments. ALPS fulfills these criteria, taking roughly 15 minutes to administer,
including self-paced participant breaks, which makes it a suitable task to include in a
battery of other tasks that make up a typical hour-long experimental session. Moreover,
by using an image and word set that were both empirically normed, we minimized noise
and maximized effect size for an N That was relatively small, in contrast to measures
that rely on 100,000’s of responses (Nosek et al., 2002).
42
3. CHAPTER THREE
Perceiving valence in everyday objects
We obtain information about the world primarily through our visual system,
which works partly based on what we see and partly based on what we know already.
Without question, a significant part of the visual system is dedicated to the analysis of
low-level feature dimensions such as shape, color, and depth (Werner & Chalupa,
2004). At the same time, recent theories of object recognition propose that additional
information, such as semantic and contextual associations, contribute to visual
recognition in the form of top-down projections from the prefrontal cortex (Bar, 2007;
Bar et al., 2006). Here, we extend that framework by proposing that valence, that is, the
perceived positivity or negativity of an object, also contributes critical information to
perception and recognition.
As discussed in chapter 1, our experience of the world is not cold and cognitive;
rather we rapidly perceive and are affected by the valence of visual information (Calvo &
Lang, 2004; Greenwald et al., 1998; Klauer & Musch, 2003). We know that humans
possess highly evolved visual and affective systems that give reality to the feelings
evoked by strongly affective faces, objects, and scenes (Calvo & Avero, 2008; Fazio et
al., 1995; Lebrecht et al., 2009; Moriguchi et al., 2011; Weierich et al., 2010). However,
what about common objects? That is, objects that are typically thought of as neutral
(i.e., not having valence)? Here we propose that valences do not have to be strong or
obvious to exert an effect on perception. In fact, we hypothesize that observers
perceive subtle valences for the majority of seemingly “neutral” objects encountered in
their everyday perception; we term these valences “micro-valences”.
43
As we mentioned in chapter 1, the perception of micro-valence most likely
arises from an integration of visual properties and learned associations. We know from
previous research that observers assign valences to novel shapes and abstract patterns
(Bar & Neta, 2006; Kirk, Skov, Hulme, Christensen, & Zeki, 2009; Rentschler et al.,
1999), indicating that valence information is present in the visual properties alone;
however, observers do report more reliable ratings for real world images compared to
abstract scenes, from which we may conclude that experience based associations
contributes large amounts of information in the judgment of valence (Vessel & Rubin,
2010).
We propose that object valence operates on a continuum that ranges from
strongly positive to strongly negative (Figure 1). In the past, affect has been defined as
the combination of two continuous dimensions: valence and arousal (Colibazzi et al.,
2010; Russell, 1980; Russell & Barrett, 1999; Russell & Carroll, 1999). Because our
focus in this paper is on the micro-valences of everyday objects, we plan on restricting
our discussion to valence in this paper. Generally, researchers taking a dimensional
view of valence have not focused on the central “neutral” region, preferring instead to
study strongly affective objects that fall at either end of the continuum (Avero & Calvo,
2006; Moriguchi et al., 2011; Weierich et al., 2010). This research has been critical in
establishing a framework for affective object perception. However, the importance of
the neutral region is perhaps underestimated, because when focused on the extreme
ends, it appears that micro-valence only accounts for a small, insignificant region.
Although, when one considers that the majority of objects encountered everyday
generate a perception of valence it becomes an effect of significant magnitude.
Although valence is closely related to preference, we do not consider them to be
the same mental property or process. Valence is a single dimension that represents the
44
positivity or negativity of an object, whereas preference is defined primarily in terms of
behavior (Lichtenstein & Slovic, 2006). Which is to say, if someone selects a water glass
over a wine glass, they are assumed to have a preference for water glasses. Yet in this
case, like most cases, preference accounts for more than just the perceived valence of
an object; it also incorporates the expected reward value of the available choices given,
current goals, motivation, and context (Rangel et al., 2008). Here, if a person were
simply thirsty for water, then the selection might be made based on function rather than
valence. On the other hand, we expect valence to play a larger role in preference when
the choice comes down to objects from the same category. That is when choosing
between a tall narrow water glass and a short rounded water glass (that hold the same
volume) valence becomes a more relevant feature than function. As such, we
hypothesize that the effects of valence on perception will be most pronounced at the
subordinate level of recognition. Consequently, perceptual judgments within a given
category, for example, selecting between teapots or watches, are the most likely to be
reliably affected by the relative valences of the objects in question.
We propose that observers rapidly and automatically perceive the micro-valence
of all objects present in the visual scene, usually without awareness or a conscious
appraisal of whether they like or dislike the object. Just as we cannot help but perceive
an object’s shape and color, similarly we cannot help but perceive a valence about an
object. However, unlike shape and color, valence is a more elusive property of objects,
because is derived from multiple components, which makes the task of devising a
reliable behavioral measure of micro-valence challenging. In this chapter, we present a
task designed to capture the micro-valences of everyday objects. We then compare
this measure to a standard measure of valence typically used for strongly valenced
objects. Finally, we examine whether measured micro-valences plays a role in object
45
perception and whether these effects are more pronounced at the subordinate versus
basic level of recognition.
3.1 The “birthday” task
As aforementioned, we hypothesize that an object’s micro-valence is perceived
rapidly, often times without conscious awareness or appraisal. To capture such an
effect, we designed a task where participants did not have to report the valence of the
object directly. Instead, participants selected objects that they would most like to keep
or return had they been given as birthday gifts. The stimuli were presented rapidly, with
only a brief response window to encourage participants towards an automatic and
away from a controlled level of processing. The foremost null hypothesis was that
participants do not perceive the valence in everyday objects and so would simply select
gifts randomly. To address such a concern we designed a task that would assess the
consistency in response selection both within and across individuals. To that end, each
object was repeated multiple times in both tasks. We predicted that participants would
consistently select objects they perceived to have a positive micro-valence in the keep
task, yet rarely select them in the return task. We assume the opposite pattern of
response for objects perceived to have a negative micro-valence.
Methods
Participants
Participants were twenty individuals (ten female) from the Carnegie Mellon
University community (mean age = 23.25 years; s.d. = 3 years). Protocols were
approved and conducted in accordance with the Institutional Review Boards at
46
Carnegie Mellon University. All participants gave informed written consent and were
compensated for their time either by monetary payment or course credit.
Stimuli
The 120 stimuli used in the following experiments were colored images of non-
living visual objects, 400 x 400 pixels in size, presented onto a white screen. Each
object triplet was presented equidistance apart from the center of the screen. The
objects were randomly selected from the following ten everyday categories: armchairs,
cameras, chairs, mugs, lamps, speakers, radios, teapots, telephones, and wall clocks.
Each category contained 12 exemplars. All 120 objects can be seen in appendix 3.
The stimuli were presented using Matlab’s Psychophysics Toolbox 3 (Brainard,
1997) (Pelli, 1997) on an LG LCD monitor (1024x768 resolution) at a distance of 60 cm
from the participants.
Task
Before starting this task participants were told, “It’s your birthday! And you have
been given a series of gifts”. In the “keep” condition participants were told that on any
given trial they would see three gifts and their task was to select the gift they would
most like to keep. During each trial three objects were presented simultaneously, side-
by-side, from the same object category and the participant was instructed to select the
object that they would have most liked to have kept by making a button press to
indicate the location of the object. Each object triplet is presented for 1.5 seconds.
Participants were allowed to respond while the objects are on the screen or during a 1.5
second response window that followed. Participants were instructed to view all the
items and then make their response as quickly and as accurately as possible, based
47
only on the objects present in the current trial. Each object was repeated in unique
triplets a total of five times in this part of the experiment. The ordering of triplet
presentations was randomized across participants, but the actual object combinations
within a given triplet were the same across participants. This design allowed us to
assess consistency across participants.
The return condition was identical to the keep condition just described, but here
participants decided which of the birthday gifts presented they would most like to
return. The keep condition was designed to index the positive dimension of micro-
valence, whereas the return condition was designed to index the negative dimension.
The ordering of these two conditions was counter balanced across participants.
Results and Discussion
To calculate an object’s micro-valence from the “birthday” task, we added a
point to an object every time it was selected in the keep condition and subtracted a
point every time it was selected in the return condition. Because objects were
presented five times in each condition, the micro-valence scores ranged from -5 to +5.
By calculating micro-valence in this way, we could determine whether the participants
were selecting objects randomly, that is, picking gifts indiscriminately in both the keep
and the return task, which would result in a micro-valence score of zero, or close to
zero; or alternatively, whether participants were picking gifts consistently in one
condition and not the other. For example, if an object were selected four times on the
keep condition, then never selected on the return condition, it would have an overall
valence of +4. Likewise, if an object were selected four times on the return condition
and never picked on the keep condition, it would have an overall micro-valence of -4.
48
Results indicate that individuals are consistent with respect to their selections.
Of the twenty participants we ran, a large number of them had micro-valence scores of
(+/-) 4 or (+/-) 5 demonstrating high levels of consistency (for details, see Table. 3.1).
Part of the reason we divided this experiment into the keep condition and the return
condition (and counter-balanced the ordering of tasks across participants) was to
control for repetition priming. If participants were simply picking objects based only on
their previous selection, then we would expect participants to continue picking the
same objects across both tasks. This being the case, the resulting micro-valence
scores would be close to zero, which is not what we observe (Figure 3.1).
Figure 3.1 presents histograms of micro-valence scores generated by each individual in the “birthday” task. The horizontal axis represents the score for a given object that ranges from -5 to +5, and the vertical axis represents the number of participants that rated the object with the particular micro-valence score. Histograms colored in red indicate objects with a distribution significantly skewed towards positive, whereas histograms colored in blue indicate objects with a distribution significantly skewed towards negative. Significance was measured by distributions where the 95% confidence intervals for the distribution did not span zero.
!
"
#
$
%
!"#$%&'( )*+#*
,($-"#".(+-/
&'
!
"
#
$
%
!"#$%&'( )*+#*
,($-"#".(+-/
&'
!
"
#
$
%
!"#$%&'( )*+#*
!"#$%&%'"($)
&'
!
"
#
$
%
!"#$%&'( )*+#*
()*+,-,.)/+0
&'
!
"
#
$
%
!"#$%&'( )*+#*
!"#$%&%'"($)
&'
!
"
#
$
%
!"#$%&'( )*+#*
()*+,-,.)/+0
&'
49
Micro-valence -5 -4 +4 +5
Proportion of Participants 17/20 18/20 19/20 17/20
Table 3.1 reports the proportion of participants that were consistent in their assignment of micro-valence scores in the “birthday” task. The table shows the number of participants out of a total of 20 that perceived at least one object to have a micro-valence score of -5, -4, +4, or +5.
Another potential account for the consistency in object selection comes from
the Free Choice Paradigm (Brehm, 1956; Harmon-Jones & Mills, 1999; Sharot, De
Martino, & Dolan, 2009). These experiments demonstrate that participants who at first
rate two objects as equivalent, go on to rate subsequently chosen items as more
positive following a forced choice selection. Theorists speculate this change arises from
feelings of cognitive dissonance (Festinger, 1957), meaning that people strengthen their
preferences to avoid feeling like they have made the wrong choice. We argue that this
theory is unlikely to account for our present results. As Figure 3.1 illustrates, the micro-
valence ratings from the birthday task show surprising levels of consistency across
participants. For the cognitive dissonance theory to explain this consistency then all
participants would have had to both see and select the same objects in the first few
trials. Given the randomization of trial order across participants this could not have
happened, which means that the free choice paradigm cannot explain the consistency
across participants in micro-valence ratings that we obtained in this experiment.
Finally, we should note that the group averages depicted in the histograms in
Figure 3.1 have a distribution that is significantly skewed (i.e. the confidence intervals
for the distribution do not span zero), and of the 120 objects used in the experiment, 52
50
are significantly skewed. 26 are skewed in a positive direction, and 26 are skewed in a
negative direction. This surprising consensus in micro-valence ratings could be
explained in two ways. Micro-valence may be a subjective construct, purely a
reconstruction from an individual’s affective associations. Conversely, micro-valence
may be more objective in nature, a property contained within an object’s physical
properties (such as color and shape). At this point we speculate that some combination
of the two would accurately capture the micro-valences we report. We do, however,
acknowledge that our population is a fairly homogenous group of undergraduates who
likely have similar life experiences, which may account, in part, for the consensus we
see here.
3.2 The ranking task
The micro-valence scores reported in the birthday task are derived from choices
or selections and not a direct measure of valence per se, so it is important to compare
our results to a task that directly measures valence. As such, we devised a task
whereby participants were required to rank order objects from most negative to most
positive along the dimension of valence. Unlike the birthday task, here participants were
explicitly evaluating the valence of each object. That being said, in an attempt to guide
participants towards a more automatic processing mode, we did set a 3-minute timer
on each trial and instructed participants to order the objects as quickly as possible
based on their initial “gut” assessment of valence.
Methods
Participants and Stimuli identical to Experiment 3.1.
51
Task
In this task, on a single trial, participants were presented with thumbnail images,
65x65 pixels in size, of all twelve exemplars from a given category. At the beginning of
each trial, the objects were randomly assigned to a position on the screen. The
participants’ task was to rank the objects from left to right, with far left being the most
negative and far right being the most positive. Scrolling over the thumbnail with the
cursor enlarged the image to 400x400 pixels. Subjects used the cursor to drag the
object to their desired position on screen. Participants were given three minutes to
complete this task. 30 seconds before their time elapsed, a stopwatch timer was
presented in the top left hand corner to indicate the time remaining. There was a single
trial for each of the ten object categories.
Results and Discussion
For every trial we recorded the x-y screen pixel co-ordinates for each object
from which we were able to assign the ranked position. The object in the far left (most
negative) position on the screen was assigned a 1, and the object in the far right (most
positive) position was assigned a 12. Because each object category formed its own
trial, then each object within a category was assigned a number from 1-12. The ranked
position number for each object was then correlated with the corresponding score from
the “birthday” task derived in Experiment 3.1.
In that we observed reasonably strong consensus for micro-valence in the
“birthday” task, for each object we used the group averages to compute a correlation
between the “birthday” task and the ranking task. Our objective was to determine
whether the scores from a task that measured valence directly predicted scores from
the more implicit, choice-based measure of the birthday task. The results in Figure 3.2
52
show a significant positive correlation between the two tasks [r2 = 0.76, p = .0001], and
there are no differences between the particular categories used in the experiment, as
indicated by the spread of data points.
Figure 3.2 shows a significant linear correlation between the group average scores from the ranking task and the group average scores from the “birthday” task [r2 = 0.76, p = .0001]. Each dot represents the average ratings for a single object on both tasks. Dots are color coded according to their basic level category. The distribution of colored dots indicates the correlation is not driven by a preference for a particular object category.
In order to confirm that the reported correlation was not driven by either (a) a
particular object category, or (b) by a small subset of participants, we created a “heat
map” that depicts the r-values for the birthday task – rank task correlation for each
object category, for each participant (Figure 3.3). As indicated by the large red areas on
-3!
-2!
-1!
0!
1!
2!
3!
0! 2! 4! 6! 8! 10! 12!
Avera
ge s
co
re o
n "
birth
day"
task!
Average score on the ranking task!
camera!
chair!
arm chair!
lamp!
mug!
radio!
speaker!
teapot!
telephone!
clock!
53
the heat map, the majority of participants showed strong correlations between the two
measures. This significant correlation confirms that the dimension of valence drives a
large part of the selection of objects in the “birthday” task, and the corresponding
micro-valence score.
Figure 3.3 presents the r-values for the correlation between the “birthday” task and the ranking task for each participant for each object category. Each row of the table represents the r-values for a single participant across the different categories. Each column represents the r-values for different participants for the same object category. Cells colored red indicate a high positive correlation between the two tasks, cells colored blue represent a high negative correlation, and darker cells indicate a low correlation.
Finally, the high correlation between the “birthday” task and the ranking task is
supported by another more direct measure of valence that we collected. In this task,
participants clicked a point on a line that represented the valence scale from positive to
!"#!$!#%&'()&*+!,!)&"+-
./
.0
.1
.2
.3
.4
.//
./5
.6
.7
.8
./0
./3
./4
./1
./6
.05
.0/
./2
./8
./7
54
negative. When taking the average object ratings and correlating them with the average
birthday task we report another significant correlation [r2 = 0.51, p = .0001], providing
additional corroborative evidence that scores on the ‘birthday” task are capturing
something about the valence of an object.
3.3 How micro-valence influences object recognition
Experiment 3.1 establishes that participants perceive the micro-valences in
everyday objects. Here, in Experiment 3.3, we examine the effects of such valences on
object perception. Because participants can perceive valence in categories typically
regarded as non-affective, such as lampshades and telephones, it is highly likely that
the mechanisms responsible for perceiving such valences are online constantly and
form part of our general object perception. We propose valence operates as a feature
dimension, in much the same way as shape, color, and semantics, which means that
we predict that valence is encoded during the early stages of object perception and not
a post-hoc label applied after recognition.
Even if micro-valence is perceived for the majority of objects present in the
visual scene, it may exert more influence over perception and behavior at different
levels of recognition. As previously discussed, when selecting between two objects
from different categories, with different functions, valence likely provides a less relevant
cue than when selecting between two objects from within the same object category.
Therefore, we designed a simultaneous identity-matching task to address the role of
valence in object perception at both the basic and the subordinate levels of visual
recognition. We predicted that on trials where the objects were qualitatively matched in
valence, participants would be slower to make a same/difference judgment, because
similarity in valence is assumed to render the objects harder to discern even if the visual
55
features are different. Critically, we also predicted that these effects would be much
more pronounced at the subordinate level.
Methods:
Participants
Participants were twenty individuals (eleven female) from the Carnegie Mellon
University community (mean age = 23.35 years; s.d. = 3.66 years). Protocols were
approved and conducted in accordance with the Institutional Review Boards at
Carnegie Mellon University. All participants gave informed written consent and were
compensated for their time by monetary payment.
Stimuli
The 64 stimuli used in the following experiment were colored images of non-
living visual objects, 400 x 400 pixels in size, presented in the center of a white screen.
For the main experiment, 47 objects of the total 64 used were selected from the
following four everyday categories: armchairs, cameras, teapots, and telephones used
in Experiment 1. These particular categories were selected because they contained
roughly equal numbers of positive and negative objects, and because they formed four
fairly distinct object categories. The micro-valence for these objects ranged from -2.82
to +2.58. These ratings were calculated by taking an average of the individual ratings
from the “birthday” experiment in Experiment 3.1. Each category contained twelve
exemplars, with the exception of the camera category where we dropped one exemplar
that had an average micro-valence of zero.
56
For the practice trials, we selected 17 objects from the categories chairs and
lamps. These objects were also from the initial object set used in Experiment 3.1. None
of these images appeared in the experimental trials for this task.
The stimuli were presented using Matlab’s Psychophysics Toolbox 3 (Brainard,
1997; Pelli, 1997) on an LG LCD monitor (1024x768 resolution) at a distance of 60 cm
from the participants.
Task
On any given trial the participant was presented with an object in the center of
the screen for 17ms after which a white noise mask replaced the object for 500ms. A
second object then appeared for 17ms, during which time the participant responded as
to whether the object was the same (i.e. an identical object), or different. There was
then a 500ms inter-trial interval (ITI) before the onset of the next trial.
There were eight conditions in this experiment where micro-valence (positive
versus negative) was fully crossed with object level (basic versus subordinate):
Positive/Positive (PP), Negative/Negative (NN), Positive/Negative (PN), and
Negative/Positive (NP), in which the first letter corresponds to the valence of the first
object, and the second letter to the second object, each of these conditions repeated
for basic and subordinate object pairings. Objects were defined as positive or negative
based on the average ratings from the “birthday” task (Experiment. 3.1); note that the
participants that generated the micro-valence ratings in the birthday task were different
than the participants in this current task. To create identical match trials, needed for
participants to make an identity same/different judgment, each of the 47 objects
appeared as both the first and the second object on one trial. Before the participant
57
began the experiment, they underwent ten practice trials to familiarize themselves with
the task and the speeded presentation times.
Results and Discussion
Response times that exceeded 2 standard deviations from the mean were
identified and excluded from subsequent analyses, as were same trials (where the same
object appeared on both presentations).
The average response times for (PP & NN) and (NP & PN) showed the same
pattern and so for simplicity we averaged across the conditions and report effects for
congruent valence (PP & NN) and incongruent valence (NP & PN) for both the basic and
the subordinate level. Figure 3.4 shows that the effects of congruent valence are
different at the basic and subordinate levels confirmed by the significant interaction of
congruency x object level [F (1, 19) = 17.72, p = .0005]. There is no significant main
effect of valence, but we do report a significant effect of object level [F (1, 19) =
11.0395, p = .0036], demonstrating that the subordinate task is slightly harder than the
basic level task, which is to be expected. The pattern of results in Figure 3.3 shows that
at the basic level participants are significantly faster to respond on valence congruent
(compared to incongruent) trials [t (19) = 2.306, p = .0325]. Yet in contrast, at the
subordinate level, participants are faster to respond on the valence incongruent
(compared to congruent) trials [t (19) = -3.64, .0017]. These results indicate a complex
interaction between valence and recognition speed, which we will attempt to explain
below.
58
Figure 3.4 presents data from the same/different identity-matching task (exp. 3.3). Results confirm that participants are faster to respond during valence congruent trials at the basic level [t (19) = 2.306, p = .0325] depicted by the columns on the left hand side of the graph, yet they are slower to respond during valence congruent trials at the subordinate level [t (19) = -3.64, .0017] presented on the right hand side.
Our current theory proposes that people perceive valence in the same way they
do other features, such as, shape, color, and semantics (see Figure 3.4). With respect
to the current task, this implies that participants automatically perceive the valence in
each of the objects, regardless of object category, or trial type; in other words, if they
have perceived the object then they have perceived the object’s valence. The
differential response times for basic and subordinate trials come from how this
automatically generated valence interacts with the demands of the task.
We know from affective priming studies that response times are facilitated
during trials where two stimuli are matched in valence (see discussions in chapter 2).
Given that our experiment included some sequential trials where the two objects shared
the same valence, then we would expect some level of affective priming in our data,
even if other more dominant processing masks it.
!"#$% !"#$%&$'"()
*+,
-.,
-/,
012)% (34)5) 6
7839#
:" 6)'%)3;<'=>?)'(
:" 6)'%)3@'%<'=>?)'(
59
We do actually report a speeded response time for congruent versus
incongruent trials at the basic level, suggestive of affective priming. Although present in
the perception of the object, valence is not really a necessary feature for deciding
whether two basic level objects are the same or different. A chair and a teapot differ on
more salient dimensions than valence, which means that it probably does not
contribute critically to performance on the task. Nonetheless it is present and is
detectable by the slight (10ms average) response time advantage for congruent trials,
indicative of some level of affective priming.
The mechanisms underlying affective priming are so automatic that such
priming would also occur in the subordinate level comparisons; however, at this level
the task demands are slightly different (shown by the main effect of object level), which
means that other, more dominant processing may be masking any effects of affective
priming. We should note that the reported findings from the subordinate level are
consistent with our prediction that participants take longer to report differences
between objects when the valence is congruent versus incongruent. We hypothesize
that this occurs because at the subordinate level many object features are shared
across the comparison stimuli, so valence becomes a critical feature dimension used
for differentiation. Trials containing two objects matched along this dimension will be
more confusable and harder to differentiate, explaining why we report longer response
times on valence congruent trials at the subordinate level only.
It appears that there are three different levels of processing operating in this
task. Firstly, participants automatically perceive the object’s valence during recognition.
Secondly, if the two objects on a given trial match in valence, affective priming occurs.
And finally, if valence is used as a critical feature for differentiation, then during trials
where the valence is matched, the perceptual task becomes more difficult. The critical
60
point to note is that because these results show differential effects of micro-valence on
object recognition, then valence must be some operating as some feature of object
perception and not as an affective label assigned after recognition takes place.
General Discussion
This chapter is predicated on the idea that, during perception, everyday
seemingly “neutral” objects are perceived to have a subtle valence that we refer to as a
micro-valence. The objects in our study come from nominally non-affective categories,
such as wall clocks, lamps, armchairs, teacups etc. And yet, across several tasks we
were successful in demonstrating that participants perceive the micro-valence in
everyday objects, and its effects are present right down to the level of object
recognition.
To review, in Experiment 3.1, we devised the “birthday” task, a novel paradigm
designed to capture the automatic perception of micro-valence. Using this task we
were able to show that participants (a) perceived micro-valence in objects, (b) were
consistent in their perceptions, and (c) converged on group averages that convey high
levels of consensus. This experiment was conducted using 120 randomly selected
everyday objects, but we assume these findings extend to the majority of objects seen
in during everyday perception.
We acknowledge that the “birthday” task relies on a choice or preference based
task to index micro-valence. However, in Experiment 3.2 we demonstrate that these
ratings are driven in large part by valence, based on a significant correlation between
the scores from the “birthday” task and scores from a direct measure of valence.
Critically, we proposed that valence is a dimension of perception, computed for
the majority of objects, even ones thought previously to be non-affective or neutral.
61
Based on this prediction we ran Experiment 3.3 to examine how micro-valence impacts
performance in visual object recognition. We used a simultaneous object-matching
task, a common standard for the field of object recognition (Tanaka & Taylor, 1991), in
which we find valence can influence recognition judgments differentially for basic and
subordinate levels.
These results force us to reassess our understanding of the contribution of
valence to everyday perceptions and behaviors. Previously, people have regarded
valence to only exert an effect when the valence is strong and clearly pronounced
(Colibazzi et al., 2010). However, we have shown here that even the smallest valence
attributed to a teapot can affect both object recognition and selection. Based on these
findings we suggest that with respect to objects, valence can be defined as another
feature that is computed automatically for the majority of objects we perceive in our
everyday environment.
Figure 3.5 provides a schematic illustrating how valence can be regarded as an object property. The example object here is a teapot to illustrate how valence is not just a feature of objects with a strong and pronounced valence, but is also potentially a property of all everyday, seemingly neutral objects.
function
material
related objects
graspvalence
depth
color
62
This is in contrast to traditional models that propose 1) a complete object
representation is generated by the visual system then 2) the affective system assigns a
label only when the object is deemed to be affective enough (Grabenhorst & Rolls,
2011). This leaves the very open question of how the affective system knows to assign
a particular object with a particular affective label. How does the affective system sort
through all of the incoming visual information to ensure it does not miss a critical object
in the environment? Our model proposes that if valence operates as a dimension of
object recognition, then all incoming visual information will be assigned a valence,
regardless of strength value. Having this mechanism in place means that if a visual
stimulus exceeds a valence threshold (i.e. is strongly positive or strongly negative) then
a signal can be sent to the affective system for it to generate a complete affective
response. If not, then the valence metric can be used for choice selections or object
recognition as we demonstrated in the experiments in this chapter.
63
4. CHAPTER FOUR
The neural basis of object recognition
Introduction
Perception extends beyond perception per se. High-level inference constantly
informs the analysis of low-level visual information. Knowledge, experience, and
feelings not immediately present in the visual field participate in perception (Barrett &
Bar, 2009). This type of information can be represented in the form of associations (Bar,
2007) that are reactivated during automatic memory retrieval (Neely, 1977). Recent
research indicates that these object-based associations are encoded in the prefrontal
cortex and that they facilitate visual recognition via top-down projections to ventral
temporal regions critical to object processing (Bar et al., 2006). This finding and findings
like this, force us to reconsider the scope of what constitutes visual perception, in
particular, thinking more broadly about neural systems may be contributing to object
perception. Put another way, it is becoming increasingly clear that visual object
recognition cannot be construed as a bottom-up hierarchical process (Marr, 1982)
The majority of research addressing top-down inputs to visual recognition has
been mostly restricted to semantic associations (Bar, 2007; Barrett & Bar, 2009;
Gauthier, James, Curby, & Tarr, 2003; James & Gauthier, 2003) is one of the few
exceptions). In the current chapter we extend those findings to investigate to what
extent valence-based associations contribute to the recognition of objects. We predict
that they will contribute to perception in a similar way to semantic associations, that is,
via feedback projections from PFC to ventral temporal object regions.
64
In chapter 3, we established that the perception of valence is continuous rather
than categorical, based on our finding that everyday objects, which one might have
been expected to be categorized as neutral, instead gave rise to the perception of
micro-valence. In this chapter then our aims are twofold. First, we plan to build on the
findings that object valence operates along a continuum. Until now, our main evidence
has been that objects in the central region of the continuum are not neutral, but we
have not actually demonstrated that valence is coded along a continuum with respect
to the entire range (strongly negative to strongly positive). To address this, we will use
functional neuroimaging to explore whether the intensity of a perceived valence
corresponds to the intensity in neural response in brain regions sensitive to valence. If
we find this to be the case, this constitutes some evidence that object valence indeed
operates along a continuum.
Our second objective relates to the prediction that valence is not just a
continuum or dimension, but more specifically, is a feature of perception. We have
already shown that valence can differentially influence perceptual judgments at the
basic and subordinate levels of object recognition; strongly suggestive that valence is
involved in perception. Here we aim to strengthen our understanding of how valence
contributes to perception by looking for effects of valence within the network of brain
regions implicated in object recognition. That is, we know that top-down signals are
likely to provide information that facilitates visual recognition, thus, it is reasonable to
posit that effects of this top-down information should manifest in visual areas. As such,
this study examines neural responses both in regions of prefrontal cortex (PFC), and in
the lateral occipital cortex (LOC), a component of the ventral-temporal cortex known to
be critical for object recognition (Grill-Spector, Kourtzi, & Kanwisher, 2001).
65
In an effort to understand how micro-valence relates to what is already
understood about object valence, we needed to identify the neural network involved in
processing the valence of strongly affective objects. By understanding where objects
with a strong and obvious valence are represented we can then plausibly understand:
a) whether micro-valences are represented in the same location and so arguably
processed by the same system; b) whether micro-valences are coded along a
continuum that ranges from strongly positive, through micro positive and micro
negative, to strongly negative. In order to address these issues, in this fMRI study we
crossed valence (positive or negative) with strength (strong or micro) to yield the four
following conditions: strong positive (SP), strong negative (SN), micro positive (MP), and
micro negative (MN).
Of procedural note, whilst collecting pilot data in preparation for this experiment
we determined that the perception of micro-valence is sensitive to the context in which
the object is presented. More specifically, in collecting valence norm ratings for
everyday objects we used two different conditions: one in which everyday objects
appeared with other objects with strong valences and another in which everyday
objects appeared alone. Our results suggest that participants consistently
underestimated the micro-valence of objects if they were intermixed with highly
affective objects in the same experiment. It is not surprising that a teacup is perceived
as less positive following the presentation of a bundle of $50 bills. Given these pilot
results we structured our experiment so that all micro-valence objects appeared in the
first five runs and so were necessarily independent from anchoring effects due to the
stronger objects that appeared in the last five runs.
In an effort to localize the neural network coding for object valence as it relates
to perception we ran two independent localizer scans. The first was a standard object
66
localizer (objects minus their phase scrambled counter parts), designed to localize the
regions involved in object perception (e.g., LOC). The second was an affective localizer
(positive and negative objects with a strong valence, minus those with a neutral
valence) designed to localize the regions involved in processing affective valence.
Participants
Participants were fifteen right-handed individuals (five male) from the Brown
University community between the ages of 18 and 35 with normal or corrected to
normal vision and screened for contraindications for MRI. Protocols were approved and
conducted in accordance with the Institutional Review Board at Brown University. All
participants gave informed written consent and were compensated for their time. Four
participants were removed from the analysis due to excessive head motion.
Stimuli
Experiment
Of the 240 non-living objects used in the experiment, 120 objects were
considered to have a micro-valence and 120 to have a strong valence. The micro-
valence objects were selected from the following ten everyday categories: armchairs,
cameras, chairs, lamps, mugs, radios, speakers, teapots, telephones, and wall clocks.
The objects with a strong valence were selected from a variety of different categories,
examples of which were: flowers, cakes, weapons, and trash. All objects were assigned
to either the positive or negative category based on behavioral norms, collected from
an independent group of participants. For the strong objects the norming consisted of
rating an object on a valence scale that ranged from 0-1 (negative to positive) by
clicking a point on a line underneath the object that indicated the valence. Objects were
67
assigned a negative valence if the average rating was between 0.0 and 0.4, and to the
positive if it was between 0.6 and 1.0. Objects that did not reach these criteria were
removed from the stimulus set and not used in subsequent experiments. The final
stimulus set included 60 positive objects and 60 negative objects.
Objects with a micro-valence were assigned either a positive or negative
valence based on scores from the “birthday” task (for details, see chapter 3) run prior to
the current experiment with an independent group of participants. Objects assigned a
positive micro-valence were those that had a “birthday” score of above zero.
Alternatively, objects with a “birthday” score of below zero were assigned a negative
micro-valence. Using this criterion, of the 120 micro objects used in the experiment, 57
were micro-positive and 63 were micro-negative. See appendix for all of the images
used.
Localizers:
Object Localizer: The total 192 objects consisted of 96 everyday color object images
and their phase scrambled counter parts. The objects were both living and non-living
and not controlled for with respect to valence.
Affect Localizer: The total 192 objects consisted of 96 objects normalized for valence,
resulting in 48 positive objects and 48 negative objects (norming procedures and cut off
criterion was identical to the experimental stimuli). The remaining 96 objects were
considered to be as close to neutral in valence as was possible. Example objects
included paperclips, blank CD cases, plastic containers etc.
Procedure
Experiment
68
Whilst in the scanner participants were presented with a single object for 500ms
in the center of a white screen and asked to rate it for pleasantness on a 1-4 scale
using a response box. Participants were able to respond while the object was on the
screen, or during a 1500ms response window that followed. Trials were separated by a
12 second ITI, during which time participants focused on a central red fixation cross.
fMRI procedure
The ten experimental runs were organized such that micro-valence objects were
presented in runs 1-5 and objects with a strong valence were presented in runs 6-10.
Within a given run the presentation of positive and negative objects was randomized.
Participants were given the task instructions outside of the magnet, but at the
beginning of each run an instruction screen was presented for 10 seconds as a
reminder. The instruction screen was followed by 10 seconds of fixation before the
onset of the first trial. There were 24 trials per run and each run lasted approximately
5 minutes.
Localizer
The object and affect localizer were identical with the only exception being the
stimuli that were presented to participants. For each localizer there was 1 run that
contained 12 sixteen-second blocks separated by 6 seconds of fixation. Single objects
were presented in the center of a white screen, during which time participants were
instructed to look for an identical object match based on the preceding or upcoming
object. The object localizer always preceded the affect localizer, and the experimental
runs always preceded both localizer scans.
69
fMRI procedure
Whole brain imaging was performed on a Siemens 3.0 T TIM Trio MRI Scanner.
At the start of the scan session high-resolution T1-weighted (magnetization-prepared
rapid-acquisition gradient echo) anatomical images were collected [TR, 1900 ms; TE,
2.98 s; flip angle 9o; 160 sagittal slices 1x1x1mm]. The following 10 experimental runs,
and 2 localizer runs were acquired using a gradient-echo echoplanar sequence
[repetition time (TR), 2 secs; echo time (TE), 30 ms; flip angle, 90o; 40 slices; 3x3x3mm].
Stimuli were presented on an Apple Macintosh computer and displayed on a rear
projection system via a mirror attached to a 32-channel head coil. Manual responses
were collected using a Mag Design and Engineering four-button response pad and
recorded using Psychophysical Toolbox (Brainard, 1997; Kleiner et al., 2007) running
within Matlab.
fMRI data analysis
Preprocessing and data analysis were performed using SPM5
(http://www.fil.ion.ucl.ac.uk/spm/). During preprocessing stages, functional images
were corrected for differences in slice time acquisition by resampling all slices to match
the first slice. Using sinc interpolation, images were motion corrected across all runs.
The functional data was then normalized based on the Montreal Neurological Institute
stereotaxic space and smoothed with an 8mm full-width at half-maximum isotropic
Gaussian kernel. Data analysis was conducted under the assumptions of SPM5 general
linear model. A single model was constructed that contained all of the experimental
trials, regardless of the participants’ response. The four regressors used in the model
came from the four experimental conditions, strongly positive (SP), strongly negative
(SN), micro positive (MP), and micro negative (MN). We also included a single regressor
70
for the 10 second instruction screen that repeated at the start of every run. The
regressors were generated by convolving the canonical hemodynamic response
functional with its temporal derivative for each epoch.
Contrast overlays were created using the SPM surfrend toolbox
(http://spmsurfrend.sourceforge.net/), and region of interest analysis were conducted
using the SPM marsbar toolbox (http://marsbar.sourceforge.net/). Antatomical regions
of interest were drawn using MRICRON
(http://www.cabiatl.com/mricro/mricron/index.html).
Behavioral Results
The behavioral results recorded from inside the scanner demonstrate that
participants assign pleasantness ratings faster for positive as compared to negative
objects, a result depicted in Figure 4.1 and confirmed by a main effect of valence [F(1,
15) = 27.73, p <.0001]. This finding supports the notion that observers respond more
quickly when processing positive items (Unkelbach et al., 2008; Unkelbach et al., 2010),
which we discussed in some detail in chapter 2. The results also indicate that
participants respond faster overall for objects with a strong as compared to micro-
valence, illustrated by a main effect of valence strength [F (1, 15) = 12.39, p = .003]. The
interaction between valence and valence strength is also significant [F (1, 15) = 8.43, p
= .01]; notably, the reported response time advantage for positive objects occurs for
both levels of valence strength. Participants were significantly faster for positive
compared to negative strong objects [t (15) = -4.07, p = .001]. And critically, the same
pattern is evident for the micro objects [t (15)= -3.49, p = .003]. This finding reassures
us that even though the micro stimuli were normed by an independent group of
participants, the valence ratings transfer such that the current participants evidently
71
perceive the micro-positive objects to be more positive than the micro-negative
objects.
Figure 4.1 presents the behavioral data collected during the scan session, in which participants were required to rate the objects for pleasantness on a 1-4 scale. The response times indicate that participants were significantly faster to evaluate positive objects for pleasantness than they were for negative. This is true for both objects with a strong valence [t (15) = -4.07, p = .001] and also for those with a micro valence [t (15)= -3.49, p = .003]. Overall participants were faster to respond on strong valence trials as indicated by the response times presented in the figure.
The neural representation of valence and micro-valence
In order to identify the regions involved in processing objects with a strong
valence we subtracted the activation on strong negative trials from strong positive trials
(SP-SN). As expected, we see that valence information is coded in PFC, indicated by
the activation plotted in Figure 4.2a. There are two notable clusters, one located in the
inferior frontal sulcus, and the other located in a more dorsal portion of frontal polar
cortex. The location of this activation supports our prediction that the representation of
valence contributes to object perception via top down projections from PFC.
The most critical finding is that when we repeat the same contrast for micro-
valence (MP-MN) we see activation located in an adjacent brain region. The fact we
report any activation for this contrast is by some measures surprising. We are
Micro Strong1000
1100
1200
1300
1400
RT
ms
Positive
Negative
72
contrasting valence for everyday objects that would have previously been assigned to a
neutral condition. As indicated by Figure 4.2b not only do we report activation, but also
this activation falls adjacent to the location of the strong valence condition. Given this
finding we surmise that micro-valence is coded by the same neural system that codes
for objects with a strong valence.
Figure 4.2 presents activation threshold at p <.05 for SP-SN in green (a) and MP-MN in purple (b) plotted on an inflated left hemisphere. The yellow box highlights the adjacency of activation for the strong and micro conditions in the inferior frontal sulcus and the orange box highlights a similar spatial relationship in a slightly more dorsal region of PFC.
What is interesting about these findings is not the precise location within PFC,
but rather the relative proximity between the strong and micro valence activations.
However, we should point out that with the current number of participants in this study,
these reported contrasts are uncorrected for multiple comparisons (e.g., using FDR
correction) and have a significance value of p <.05 at the level of whole brain. At the
same time, the results within each participant do reach significance at p <.001. We
posit that this is because the actual location of valence coding varies slightly across
73
participants and the group analyses blur our effects. There are two critical points to
note about this. First, the reported activation for valence is always coded in some
region of PFC. And second, within a participant, regardless of the location, the
activation for micro-valence is always clustered in an adjacent region to strong valence.
This is similar to findings that show an interesting relationship between the locations of
the face and voice area (von Kriegstein, Kleinschmidt, Sterzer, & Giraud, 2005). Table
4.1 presents the peak activation for strong and micro-valence for each individual
participants.
Table 4.1 reports the peak X, Y, and Z MNI co-ordinates for clusters from the following two contrasts: (SP-SN) which is reported as strong in the table and (MP-MN) which is reported as
Participant X Y Z Size t-value Valence Strength
S1 -36 54 30 14 4.02 strong
S1 -45 45 24 27 4.46 micro
S2 -3 63 27 147 5.5 strong
S2 -18 63 30 164 5.77 micro
S3 -33 66 12 683 5.54 strong
S3 -18 18 -9 12 4.25 micro
S4 3 63 30 16 4.66 strong
S4 6 63 12 6 3.31 micro
S5 -42 51 9 1013 9.08 strong
S5 n/a n/a n/a n/a n/a micro
S6 -18 51 15 29089 11.38 strong
S6 15 57 27 6 3.41 micro
S7 -15 69 21 132 6.24 strong
S7 -33 48 33 141 5.66 micro
S8 -18 57 6 7028 n/a strong
S8 -27 60 24 355 6.15 micro
S9 9 54 -3 371 5.09 strong
S9 -33 63 21 12 3.8 micro
S10 -6 60 3 89 4.05 strong
S10 27 63 -9 5 3.41 micro
S11 -18 66 -6 8 4.1 strong
S11 -42 54 -3 10 4.08 micro
S12 n/a n/a n/a n/a n/a strong
S12 -45 42 6 44 3.67 micro
S13 -18 63 21 245 6.32 strong
S13 -30 42 -18 42 4.9 micro
S14 -3 63 18 18 3.54 strong
S14 -18 15 -30 7 3.87 micro
S15 n/a n/a n/a n/a n/a strong
S15 -3 51 -9 55 4.05 micro
74
micro in the table significant at p < .001. For each cluster we also report the size and t-value. In cases where the participants show either no clusters, or no identifiable clusters we reported n/a. The two clusters presented in red indicate that they are not significant at p <.001 uncorrected.
Activation at the center of continuum
To highlight whether the area of activation seen in the left inferior frontal sulcus
is representing valence on a continuum we ran the following contrast analysis. The
activation for micro-positive was subtracted from strong positive and then overlaid with
the activation from the activation from micro-negative subtracted from strong negative.
The prediction is that if this region is coding for valence on a continuum then the
difference between the above two contrasts should be overlapping. This is exactly what
is shown in the activation map depicted in Figure 4.3. Furthermore, the location of this
overlap is seen in a similar region to the activation reported in the inferior frontal sulcus
Figure 4.2.
Figure 4.3 represents the activation for the contrast SP-MP in red, overlapped with SN-MN in blue. Note the colors of the cluster in left PFC that range from blue (negative) through purple (center of continuum) to red (positive). Both contrasts are significant at p<.05.
75
Region of interest analysis
In an effort to further understand whether the intensity of perceived valence
operates on a continuum we conducted a region of interest analysis. Based on a priori
predictions that the PFC contributes to object recognition via top-down projections (Bar
et al., 2006) and OFC is engaged in value processing (Rolls & Grabenhorst, 2008), we
drew an anatomical region of interest encompassing left prefrontal cortex. The results
within this ROI indicate, as expected, a continuum of valence. The time course shown
in Figure 4.4 illustrates the strongest activation for objects with a strong positive
valence, followed by micro-positive, micro-negative, and the weakest activation for
objects with a strong negative valence. When plotting the integrated percent signal
change in Figure 4.5, we can see the linear trend, as expected from the time course that
is significant across valences and strengths [F (1, 14) = 12.88, p = .0002].
Figure 4.4 graphs the time course activation for all four experimental conditions from an anatomically defined region of interest in left prefrontal cortex. The percent signal change is noted on the vertical axis and the horizontal axis represents the progression of time. Of note is how the peak of the response of the HRF is the strongest for most positive and gets progressively less as the valence changes from strongly positive to strongly negative.
76
Figure 4.5 illustrates the significant linear trend across the four different valence strengths [F (1, 14) = 12.88, p = .0002] computed from the integrated percent signal change.
The role of valence in object perception
A final goal of this experiment was to determine the neural underpinnings of
valence processing as it relates to object recognition. In particular, the lateral occipital
cortex (LOC) is known to be a key area in the processing of objects (Grill-Spector et al.,
2001). As such, we examined whether LOC responses reflect any information pertaining
to the valence of objects. We selected a region of interest in our functionally defined
LOC (from the objects minus scrambled group map). The results in Figure 4.6 show that
neurons in the LOC do distinguish between positive and negative objects, indicated by
a main effect of Valence [F (1, 14) = 8.50, p = .001*]. Moreover, the responses in LOC
are greater for objects with a strong valence as shown by a main effect of Strength [F
(1, 14) = 7.25, p = .001*]. At the same time, there is no significant Valence x Strength
interaction.
77
In chapter 3 we argued that valence may function as just another featural
property of an object. The finding that LOC, thought to be dedicated to processing
objects and their features (Grill-Spector et al., 2001; Grill-Spector et al., 1998), is
sensitive to the valence provides strong support for this conjecture – that valence is just
one object feature among many and, as such, is part and parcel of object perception.
Figure 4.6 shows the location of the LOC cluster (a) defined from an objects-scrambled localizer (peak MNI coordinate -42, -78, 9) that is used in the region of interest analysis presented in (b). The graph of activation from the ROI shows a main effect of valence [F (1, 14) = 8.50, p = .001*], demonstrating that the LOC can distinguish between objects of different valences. This is also a main effect of strength [F (1, 14) = 7.25, p = .001*], whereby activation is stronger for objects with a stronger valence.
Discussion
The fMRI results presented here provide compelling evidence that object
valence is indeed coded on a continuum. While the contrast analysis shown in Figure
78
4.3 provides some support for this claim, undoubtedly, the most striking confirmation
comes from the region of interest analysis. Here objects that generate the perception of
strong positive valence yield the strongest positive activation. What is even more telling
is that this effect is linear across the ordered conditions, with valence becoming
increasingly less positive as we move from strong positive to strong negative (Figure 4.4
and 4.5). This neural data is the clearest indication that observers process and
represent object valence along a continuum within a single cognitive system. At the
same time, neural evidence that micro-valence is coded alongside strong valence –
where it is indisputable that valence is a feature of the objects – makes the
phenomenon of micro-valence much more tangible. That is, evidence that the micro-
valence of everyday objects is represented within the same system that represents
obvious and well-defined object valence makes it significantly harder to argue that
micro-valence is epiphenomenonal. Of course, micro-valence is a subtle feature that
may not exert obvious effects in typical laboratory studies of valence. Thus, micro-
valence has been overlooked by many researchers who have preferred instead to
construe everyday objects as being affectively neutral. In light of our current findings,
we argue that the field should no longer ignore the contribution of valence (no matter
how subtle) in object perception. That is, although the behavioral findings in chapters 2
and 3 provide a behavioral foundation for micro-valence qua micro-valence, it is the
demonstration here that micro-valence is processed by the same neural network as
strong valence that is definitive in establishing micro-valence as a part of what the field
currently understands as “valence”.
In the same way that we could not conclusively demonstrate the organization of
the valence continuum without highlighting its underlying neural representation, we
could not as definitively demonstrate the contribution of valence to object perception
79
without the reported effects in LOC (Figure 4.6). By establishing that the LOC, a
fundamental region in the “core” network for visual object recognition, is coding for
valence, we are demonstrating that valence is not simply a post-perceptual semantic
property. Instead, this finding suggests that valence is integral in the perceptual
processing of objects. What is not clear from our results, but warrants further
investigation, is how exactly the LOC represents valence. One possibility is that valence
information is extracted from objects during bottom-up processing. Alternatively, the
brain regions we identified in PFC that seem to code for valence may have projections
back to LOC. Given how top-down information facilitates visual recognition (Bar et al.,
2006), we favor the second account; although, at present either explanation is
consistent with our prediction that valence is a feature of objects that is processed by
the larger object recognition network.
It is noteworthy that our results make it clear that the amygdala is not involved in
representing the continuum of object valence. This finding may seem surprising at first,
but recent work suggests that the amygdala seems to be preferentially coding for
arousal and novelty, and not for valence (Colibazzi et al., 2010; Litt et al., 2011;
Moriguchi et al., 2011; Weierich et al., 2010). Our present fMRI results are consistent
with this finding. The only contrast in which we report reliable activation in the left
amygdala is for the strongly negative objects (greater than baseline). This finding
supports a functional view of the amygdala in which arousal is primary: the strongly
negative objects are probably the only objects in our experiment that generate a
notable level of arousal. Although the other objects generate a valence perception, we
assume they carry low levels of arousal, because they are all examples of non-living
man-made objects. In particular, our stimulus set is infinitely less arousing than the
80
International Affective Picture System (IAPS) images (Lang et al., 1997) most commonly
used in experiments that report amygdala activation.
Finally, it may also appear surprising that the valence continuum is not
represented in the OFC, but instead in a slightly more dorsal region of the brain.
However, we know that OFC is representing the reward value of objects in preparation
for decision-making (Grabenhorst & Rolls, 2011; Rangel et al., 2008; Rolls &
Grabenhorst, 2008). We argue that this is different from the valence associated with an
object. What we have highlighted from the results in this chapter is that valence is a
dimension of perception that assigns a particular positive or negative metric to
individual objects. Admittedly, the likelihood is that this valence metric contributes to
the value computed in OFC, but notably, valence should be regarded as only one
source of contributing information. When considered in this way, it makes sense that
the valence continuum is represented close to the OFC, but not directly in the OFC.
To summarize, in this chapter we report two novel findings that provide powerful
support for the central goals of the thesis. These were (a) to determine whether micro-
valence is coded by the same cognitive system that represents what is, in the field,
typically construed as object valence, and (b) to understand whether object valence
plays some role visual object processing. This experiment was successful in
demonstrating the validity of both assertions. As such, we are left with the need to
rethink what is meant by valence, how it is processed, and what role(s) it has in our
perceptual, cognitive, and affective behavior.
81
5. DISCUSSION
We started the thesis with the question, “Do people perceive “neutral” everyday
objects as having micro-valences?” Our goal was to determine (a) do these valences
exist? and (b) are they coded along a continuum ranging from strongly positive to
strongly negative? Based on the body of behavioral and neuroimaging experiments
presented in this thesis, we can say with confidence that observers do perceive micro-
valences when viewing at everyday objects. Moreover, a central goal of this thesis has
been to determine whether the perception of micro-valence operates independently of
the processing of strongly affective objects or along a continuum that encompasses all
object valences regardless of magnitude. The underlying neural representation of object
valence ( chapter 4) supports the hypothesis that valence is organized on a continuum.
Furthermore, this evidence suggests that the same cognitive and neural systems that
represent the valence of a delicious cupcake or a bloodied weapon also represents the
valence of seemingly neutral objects such as teapots and lampshades. Having
established that micro-valences exist and operate on a continuum, a number of open
questions remain, which I will address next.
5.1 What is micro-valence?
The foremost unanswered question is what conditions and experiences lead to
the perception of micro-valence? In the introduction we speculated that micro-valences
arise from affective associations reactivated during perception. This is in line with
recent theories of object recognition that propose that top-down information, in the
form of associations, assist with object processing in ventral temporal object
recognition regions (Bar, 2007; Bar et al., 2006). We speculate that objects acquire
82
these affective associations through the context in which they are experienced (Bar,
2004; Barrett & Bar, 2009). For most people, perceiving a bedroom telephone
automatically generates associations of warmth, relaxation, and homeliness that have
been formulated from the context “bedroom”. These positive associations then
contribute to the perceived positive micro-valence for bedroom style telephones. On
the other hand, office telephones may generate associations of stress, tiredness, and
challenges, based on the context “office”. Accordingly, these negative associations
contribute to the perceived negative micro-valence of office phones (Lebrecht & Tarr,
2010).
Surprisingly, participants are able to use these contextual associations to
generate the perception of micro-valence for objects they have never seen before. For
example, participants in our experiments perceived the office style phones to have a
negative micro-valence, despite having never before seen the particular images. We
propose that this is possible because participants can transfer valences from familiar to
unfamiliar objects. In this way, participants can use already established associations to
rapidly perceive the valence of never before seen objects.
This notion that valence is simply derived from an individual’s experience-based
associations is difficult to reconcile with our findings that many participants perceive
the same objects to have the same valences. One possible explanation could be that
our somewhat homogenous undergraduate participant population had many shared
experiences, and therefore shared associations and valences. One way to test this
hypothesis would be to repeat the “birthday” task from chapter 3 with a different
participant demographic and compare the average valence ratings. If the ratings
differed it would provide evidence that individuals can form collective preferences
based on the information present in their environment.
83
An alternative explanation for the reported consensus in valence ratings is that
the object features heavily contribute to the perceived valence. Research demonstrates
that participants can perceive valence in novel or abstract shapes and scenes (Bar et
al., 2006; McManus, 1980; Rentschler et al., 1999; Vessel & Rubin, 2010). In light of
these findings, when conducting pilot work for our experiments we normed a set of
abstract shapes. We did this by presenting a single abstract shape against a white
background and asking participants to click a point on a line presented underneath the
object to indicate their perception of valence, where far left indicated most negative and
far right indicated most positive. The most negative objects (based on group average
ratings) are presented in Fig. 5.1a and the most positive objects are presented in Fig.
5.1a. What is interesting is how unambiguous are the valence categories. Even without
category labels, it is easy to determine which objects are the most positive and which
objects are the most negative.
84
Figure 5.1 presents the 8 of the most negative and 8 of the most positive shapes derived from a valence norming pilot experiment we conducted. The objects grouped under (a) were rated as negative, and the objects grouped under (b) were rated as positive. Note the shape and color consistency across the two different groups.
The importance of visual properties in perceived valence was also evident when
normed the objects with a strong valence in preparation for experiment 4. The norming
procedure was the same as that described above. If the valence ratings were based
only on abstract knowledge that some particular objects are positive, and other objects
are negative, then the quality of the image depicting these concepts should not affect
the ratings. However, we found the opposite to be true: the image quality really
mattered, especially for the positive objects. In particular, we found that high quality,
high resolution photographs with good lighting (Figure 5.2b) were rated more positively
than lower quality images (5.2a) with the same semantic content. This example
(a)
(b)
85
illustrates that visual properties, in addition to semantics, can cause objects to be
perceived more positively.
Figure 5.2 illustrates three example objects that have the same semantic content, but different perceived valences. Objects in (a) were low quality images, sourced from the Internet, which were subsequently not rated as positive enough to meet the experimental criteria. The objects in (b) have an identical semantic content but were sourced from a professional photographic database (http://www.thinkstockphotos.com/). The images in (b) yielded a more positive, more consistent valence rating. Notably the only difference between these two sets of images that could account for the differences in perceived valence is the image quality.
Most likely, the perception of valence and micro-valence arises from an
interaction between intrinsic visual properties and extrinsic affective associations. How
exactly that interaction functions is still an open question. It could be that individuals
find it easier to form positive associations with objects that are perceived to have
positive features. For example, a teapot with a curved shape, and a shiny, chrome finish
might capture positive associations more readily than an asymmetric, angular, coffee
pot for a conventional Western aesthetics. With respect to memory, it could be that
positive associations are more easily encoded for an object perceived to have positive
features. Alternatively, it could be that when the features of an object are perceived to
be positive, the individual is more likely to selectively recall positive associations from
(a) (b)
86
memory. We know from the data presented in chapter 4 that valence seems to be
neurally encoded in the prefrontal cortex, which is near the neural structures devoted to
the controlled retrieval of memory (Badre & Wagner, 2007). Although, these conjectures
are highly speculative, they seem worthwhile topics to explore in future research.
More generally, we need to understand how object valence fits into the
established frameworks of cognitive neuroscience. Specifically, we need to understand
how valence relates to the field of affective neuroscience. We know that systems of
affect and emotion are implicated in the processing of valence and so we discuss the
potential contribution of arousal to our understanding of micro-valence in the next
section. Throughout the thesis we have identified some close relations with valence and
preference, which we will also try and summarize in the remaining sections.
5.2 Is micro-valence micro-affective?
The term valence is often used interchangeably with affect. On closer analysis,
researchers have shown that affect is actually comprised of two continuous,
intersecting dimensions: valence (pleasantness) and arousal (activation) (Barrett, 2006;
Russell, 1980; Russell & Carroll, 1999). Until this point, we have restricted our
discussion to the single dimension of valence; however, it is worth exploring how micro-
valence may or may not relate to arousal.
Arousal is thought of as the feelings or autonomic responses that occur when an
individual experiences a strongly affective object, event, or emotion (Russell, 1980). In
the same way that valence ranges from positive to negative, arousal ranges from high
to low. It appears that arousal is a U-shaped function of valence (Barrett et al, In Prep),
which means that objects with strong valence automatically generate high arousal (for
example, a bowl of vomit is both strongly negative and at the same time highly
87
arousing). Arousal is at its lowest in the central region of the valence continuum, which
makes sense given that we would not expect a teacup or lampshade to generate high
levels of arousal (even if they do possess a micro-valence).
At this point, we do not know whether the perception of micro-valence
generates a micro-arousal in the body. Admittedly, for strongly affective objects,
valence and arousal are so highly correlated they may appear inseparable. However, for
objects with weaker, micro-valences, the perceptual system may simply assign a metric
of valence during object recognition. If this metric is below a particular threshold it may
never actually generate an arousal response.
In support of these ideas, there is evidence to suggest that valence and arousal
are computed by separable neural structures (Colibazzi et al., 2010; Litt et al., 2011;
Moriguchi et al., 2011). Colibazzi et al. found that valence is mediated mainly by dorsal
cortical and mesolimbic structures, while arousal is computed mainly by midline and
medial temporal areas. This suggests that it may be possible to assign valence to an
object in the absence of any physiological arousal.
5.3 Valence & Preference
Although object valence is not the same as preference, the two are highly
correlated—people prefer items that have a positive valence (Bar & Neta, 2006; Bar &
Neta, 2007; Barrett, 2006). However, object preference is primarily defined in terms of
behavior (Lichtenstein & Slovic, 2006) and object valence is defined in terms of the
representation of positivity or negativity attributed to an object. The object an observer
ultimately selects, which is then defined as their preference, can actually be thought of
as “the read out behavior” at the end of a series of processing steps. This means that
before an individual actually selects an object, multiple sources of information will have
88
been integrated to inform that preference. These information sources can be derived
from the observers’ internal state, their current goals and motivations, or the current
context (Rangel et al., 2008; Sugrue et al., 2005). Once the observer has established
their immediate goals, then they can evaluate the relative object choices in order to
make an informed selection or preference (Krajbich et al., 2010). Valence likely
contributes information at this stage of preference, especially when the selection is
between objects from the same basic level category, i.e. two teacups, or two
champagne flutes. But notably valence is not the only source of information being
integrated, and so should not be thought of as synonymous with preference. Instead,
object valence should be regarded as a single dimension that captures the positivity or
negativity associated with an object representation that can inform more complex
computations such as preference, choice, or decision-making.
5.4 Model for object valence
Traditional models of affective perception assume that the visual system
recognizes an object and then the affective system assigns it a label indicating positive
or negative. Even recent papers divide sensory perception into “tier one” and affective
labeling to “tier two” (Grabenhorst & Rolls, 2011). The problem with such sequential
processing models is that they do not explain how the affective system “knows” to
assign an affective label to the object in the first place. It is almost as if these models
assume there is a homunculus deciding whether objects should be affective or not. In
contrast, what we propose is that valence is actually a featural dimension of perception,
and that all objects are automatically evaluated for valence during perception
(Figure 5.3).
89
Figure 5.3 diagrams how the valence perceived during object recognition relates to decision-making and arousal. We propose that all valence metrics generated during perception can feed forward into choice and decision-making systems, regardless of their strength value. However, only valences that exceed a particular strength magnitude are projected to the arousal system and generate a complete affective response.
By describing valence as a feature dimension of objects we postulate the
information is processed during the perception of the object and not as a label
assigned post recognition. Generally speaking there is a large amount of information
about an object, not tied to the low level features that is processed during perception.
For example, an object’s function, semantics, or category, we propose valence is
simply another one of those features. Admittedly, valence is less directly tied to the
objects low-level properties, but depth perception arises from a combination of multiple
cues and that is unquestionably regarded as perception.
!+
_
arousal
decision making
preference
value
arousal
90
For the majority of everyday objects the valence will be weak in strength and
observers will simply perceive these objects to have a micro-valence. However, by
evaluating all objects for valence, the system will be prepared to detect those with a
strong valence and generate the corresponding arousal response. For those objects
where the valence metric is not strong enough to generate an arousal response, it could
nonetheless provide input to choice and decision-making systems as described.
In conclusion, objects are not as neutral as we once thought. Even everyday
objects that are seemingly neutral are perceived to have weak or subtle valences that
we term micro-valences. Evidence generated during the course of this thesis
increasingly suggests that valences, perceived automatically during object perception,
exhibit significant, measurable, and predictable effects on our recognition and
interaction with objects.
91
BIBLIOGRAPHY
Aguado, L., García-Gutierrez, A., Castañeda, E., & Saugar, C. (2007). Effects of prime task on
affective priming by facial expressions of emotion. The Spanish Journal of Psychology, 10(2),
209-17.
Anderson, D. N. (1988). The delusion of inanimate doubles. Implications for understanding the
capgras phenomenon. The British Journal of Psychiatry, 153(5), 694.
doi:10.1192/bjp.153.5.694
Aristotle. (1988). Analytica posteriora. In D. J. Herrmann & R. Chaffin (eds.), Memory in historical
perspective: The literature before ebbinghaus. New york: Springer-Verlag.
Avero, P., & Calvo, M. G. (2006). Affective priming with pictures of emotional scenes: The role of
perceptual similarity and category relatedness. The Spanish Journal of Psychology, 9(1), 10-
8.
Badre, D., & Wagner, A. D. (2007). Left ventrolateral prefrontal cortex and the cognitive control of
memory. Neuropsychologia, 45(13), 2883-901. doi:10.1016/j.neuropsychologia.2007.06.015
Bar, M. (2004). Visual objects in context. Nature Reviews. Neuroscience, 5(8), 617-29.
doi:10.1038/nrn1476
Bar, M. (2007). The proactive brain: Using analogies and associations to generate predictions.
Trends in Cognitive Sciences, 11(7), 280-9. doi:10.1016/j.tics.2007.05.005
Bar, M., & Neta, M. (2006). Humans prefer curved visual objects. Psychological Science : A
Journal of the American Psychological Society / APS, 17(8), 645-8. doi:10.1111/j.1467-
9280.2006.01759.x
Bar, M., & Neta, M. (2007). Visual elements of subjective preference modulate amygdala
activation. Neuropsychologia, 45(10), 2191-200.
doi:10.1016/j.neuropsychologia.2007.03.008
Bar, M., Aminoff, E., & Schacter, D. L. (2008). Scenes unseen: The parahippocampal cortex
intrinsically subserves contextual associations, not scenes or places per se. J Neurosci,
28(34), 8539-44. doi:10.1523/JNEUROSCI.0987-08.2008
92
Bar, M., Kassam, K. S., Ghuman, A. S., Boshyan, J., Schmid, A. M., Schmidt, A. M., . . . Halgren,
E. (2006). Top-Down facilitation of visual recognition. Proc Natl Acad Sci U S A, 103(2), 449-
454. doi:10.1073/pnas.0507062103
Bar, M., Neta, M., & Linz, H. (2006). Very first impressions. Emotion (Washington, D.C.), 6(2),
269-78. doi:10.1037/1528-3542.6.2.269
Bargh, J. A., Chaiken, S., Govender, R., & Pratto, F. (1992). The generality of the automatic
attitude activation effect. Journal of Personality and Social Psychology, 62(6), 893-912.
Barrett, L. F. (2006). Valence is a basic building block of emotional life. Journal of Research in
Personality, 40(1), 35-55. doi:10.1016/j.jrp.2005.08.006
Barrett, L. F., & Bar, M. (2009). See it with feeling: Affective predictions during object perception.
Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences,
364(1521), 1325-34. doi:10.1098/rstb.2008.0312
Baumgarten, A. G. (1750). Aesthetica. Impens. I. C. Kleyb.
Bliss-Moreau, E., Barrett, L. F., & Wright, C. I. (2008). Individual differences in learning the
affective value of others under minimal conditions. Emotion (Washington, D.C.), 8(4), 479-93.
doi:10.1037/1528-3542.8.4.479
Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: The self-assessment manikin and the
semantic differential. Journal of Behavior Therapy and Experimental Psychiatry, 25(1), 49-59.
Bradley, M. M., & Lang, P. J. (1999). Affective norms for english words (ANEW): Stimuli,
instruction manual, and affective ratings [Technical Report C-1] (Technical Report C-1).
Gainesville, Florida. The Center for Research in Psychophysiology.
Brainard, D. H. (1997). The psychophysics toolbox. Spat Vis, 10(4), 433-6.
Brehm, J. W. (1956). Postdecision changes in the desirability of alternatives. Journal of Abnormal
and Social Psychology, 386-389.
Calvo, M. G., & Avero, P. (2008). Affective priming of emotional pictures in parafoveal vision: Left
visual field advantage. Cognitive, Affective, & Behavioral Neuroscience, 8(1), 41.
doi:10.3758/CABN.8.1.41
93
Calvo, M. G., & Lang, P. J. (2004). Gaze patterns when looking at emotional pictures:
Motivationally biased attention. Motivation and Emotion, 28(3), 221-243.
Chaiken, S., & Trope, Y. (1999). Dual-Process theories in social psychology. Guilford Press.
Colibazzi, T., Posner, J., Wang, Z., Gorman, D., Gerber, A., Yu, S., . . . Peterson, B. S. (2010).
Neural systems subserving valence and arousal during the experience of induced emotions.
Emotion (Washington, D.C.), 10(3), 377-89. doi:10.1037/a0018484
Constantino, S. M., & Daw, N. D. (2010). A closer look at choice. Nature Neuroscience, 13(10),
1153-1154.
Cuthbert, B. N., Schupp, H. T., Bradley, M. M., Birbaumer, N., & Lang, P. J. (2000). Brain
potentials in affective picture processing: Covariation with autonomic arousal and affective
report. Biol Psychol, 52(2), 95-111.
Dijksterhuis, A., & Aarts, H. (2003). On wildebeests and humans. Psychological Science, 14(1),
14. doi:10.1111/1467-9280.t01-1-01412
Ellis, H. D., & Lewis, M. B. (2001). Capgras delusion: A window on face recognition. Trends in
Cognitive Sciences, 5(4), 149-156.
Ellis, H. D., Quayle, A. H., de Pauw, K. W., Szulecka, T. K., Young, A. W., & Kolkiewicz, L. A.
(1996). Delusional misidentification of inanimate objects: A literature review and
neuropsychological analysis of cognitive deficits in two cases. Cognitive Neuropsychiatry,
1(1), 27-40.
Eslinger, P. J., & Damasio, A. R. (1985). Severe disturbance of higher cognition after bilateral
frontal lobe ablation: Patient EVR. Neurology, 35, 1731-1741.
Fazio, R. H. (2001). On the automatic activation of associated evaluations: An overview.
Cognition & Emotion, 15(2), 115-141. doi:10.1080/0269993004200024
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995). Variability in automatic
activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of
Personality and Social Psychology.
94
Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., & Kardes, F. R. (1986). On the automatic
activation of attitudes. Journal of Personality and Social Psychology, 50(2), 229-238.
Ferraro, F. R., Christopherson, K., & Douglas, J. (2006). Lexical decision task performance in
blood-fearful and spider-fearful individuals. Current Psychology, 25(2), 132-143.
Festinger, L. (1957). A theory of cognitive dissonance. In Stanford, CA: Standford UP.
Gauthier, I., James, T. W., Curby, K. M., & Tarr, M. J. (2003). The influence of conceptual
knowledge on visual discrimination. Cogn Neuropsychol, 20(3), 507-23.
doi:10.1080/02643290244000275
Ghuman, A. S., & Bar, M. (2006). The influence of nonremembered affective associations on
preference. Emotion (Washington, D.C.), 6(2), 215-23. doi:10.1037/1528-3542.6.2.215
Giner-Sorolla, R., García, M. T., & Bargh, J. A. (1999). The automatic evaluation of pictures.
Social Cognition, 17(1), 76-96.
Grabenhorst, F., & Rolls, E. T. (2011). Value, pleasure and choice in the ventral prefrontal cortex.
Trends in Cognitive Sciences, 15(2), 56-67. doi:10.1016/j.tics.2010.12.004
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. (1998). Measuring individual differences in
implicit cognition: The implicit association test. Journal of Personality and Social Psychology,
74(6), 1464-80.
Grill-Spector, K., Kourtzi, Z., & Kanwisher, N. (2001). The lateral occipital complex and its role in
object recognition. Vision Res, 41(10-11), 1409-22.
Grill-Spector, K., Kushnir, T., Hendler, T., Edelman, S., Itzchak, Y., & Malach, R. (1998). A
sequence of object-processing stages revealed by fmri in the human occipital lobe. Human
Brain Mapping, 6(4), 316-28.
Harmon-Jones, E., & Mills, J. (1999). Cognitive dissonance progress on a pivotal theory in social
psychology. In Washington, DC: Braun-Brumfield.
Haxby, J. V., Gobbini, M. I., Furey, M. L., Ishai, A., Schouten, J. L., & Pietrini, P. (2001).
Distributed and overlapping representations of faces and objects in ventral temporal cortex.
Science, 293, 2425-2430.
95
Hekkert, P. (2006). Design aesthetics: Principles of pleasure in design. Psychology Science,
48(2), 157.
Hermans, Baeyens, & Eelen. (1998). Odours as affective processing context for word evaluation:
A case of cross-modal affective priming. Cognition & Emotion, 12, 601-613.
Hermans, D., De Houwer, J., & Eelen, P. (2001). A time course analysis of the affective priming
effect. Cognition & Emotion, 15(2), 143-165. doi:10.1080/0269993004200033
Hermans, De Houwer, & Eelen. (1994). The affective priming effect: Automatic activation of
evaluative information in memory. Cognition & Emotion, 515-533.
Hill, A. B., & Kemp-Wheeler, S. M. (1989). The influence of context on lexical decision times for
emotionally aversive words. Current Psychology, 8(3), 219-227.
Hirstein, W., & Ramachandran, V. S. (1997). Capgras syndrome: A novel probe for understanding
the neural representation of the identity and familiarity of persons. Proceedings of the Royal
Society of London. Series B: Biological Sciences, 264(1380), 437.
De Houwer, Hermans, & Eelen. (1998). Affective and identity priming with episodically associated
stimuli. Cognition & Emotion, 12(2), 145-169.
James, T. W., & Gauthier, I. (2003). Auditory and action semantic features activate sensory-
specific perceptual brain regions. Curr Biol, 13(20), 1792-6.
Kirk, U., Skov, M., Hulme, O., Christensen, M. S., & Zeki, S. (2009). Modulation of aesthetic value
by semantic context: An fmri study. Neuroimage, 44(3), 1125-32.
doi:10.1016/j.neuroimage.2008.10.009
Klauer, & Musch. (2003). Affective priming: Findings and theories. In Musch & Klauer (Eds.), The
psychology of evaluation: Affective processes in cognition and emotion.
Klauer, K. C. (1997). Affective priming. European Review of Social Psychology, 8(1), 67-103.
doi:10.1080/14792779643000083
Kleiner, Brainard, & Pelli. (2007). What's new in psychtoolbox-3? [Abstract]Perception, 36(ECVP
Abstract Supplement). Retrieved from
http://www.perceptionweb.com/abstract.cgi?id=v070821
96
Kohler. (1929). Gestalt psychology. In New york: Liveright.
Krajbich, I., Armel, C., & Rangel, A. (2010). Visual fixations and the computation and comparison
of value in simple choice. Nature Neuroscience, 13(10), 1292-8. doi:18729580
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1997). International affective picture system
(IAPS): Technical manual and affective ratings. NIMH Center for the Study of Emotion and
Attention.
Lang, P. J., Bradley, M. M., Fitzsimmons, J. R., Cuthbert, B. N., Scott, J. D., Moulder, B., &
Nangia, V. (1998). Emotional arousal and activation of the visual cortex: An fmri analysis.
Psychophysiology, 35.
Lebrecht, S., & Tarr, M. (2010). Defining an object's micro-valence through implicit measures.
Journal of Vision, 10(7), 966.
Lebrecht, S., Pierce, L. J., Tarr, M. J., & Tanaka, J. W. (2009). Perceptual other-race training
reduces implicit racial bias. Plos ONE, 4(1), e4215. doi:10.1371/journal.pone.0004215
Leppänen, J. M., & Hietanen, J. K. (2004). Positive facial expressions are recognized faster than
negative facial expressions, but why? Psychol Res, 69(1-2), 22-9. doi:10.1007/s00426-003-
0157-2
Lichtenstein, S., & Slovic, P. (2006). The construction of preference. Cambridge ; New York:
Cambridge University Press.
Litt, A., Plassmann, H., Shiv, B., & Rangel, A. (2011). Dissociating valuation and saliency signals
during decision-making. Cerebral Cortex (New York, N.Y. : 1991), 21(1), 95-102.
doi:10.1093/cercor/bhq065
Mandoki, K. (2007). Everyday aesthetics : Prosaics, the play of culture and social identities.
Aldershot, England ; Burlington, VT: Ashgate.
Marr, D. (1982). Vision: A computational investigation into the human representation and
processing of visual information. San Francisco: Freeman.
Martin, A. (2007). The representation of object concepts in the brain. Annu Rev Psychol, 58, 25-
45. doi:10.1146/annurev.psych.57.102904.190143
97
Martin, A., Wiggs, C. L., Ungerleider, L. G., & Haxby, J. V. (1996). Neural correlates of category-
specific knowledge. Nature, 379(6566), 649-652.
McManus, I. C. (1980). The aesthetics of simple figures. Br J Psychol, 71(Pt 4), 505-24.
Meyer, E., & Schvaneveldt, W. (1971). Facilitation in recognizing pairs of words. . Journal of
Experimental Psychology, 90, 227-234.
Moriguchi, Y., Negreira, A., Weierich, M., Dautoff, R., Dickerson, B. C., Wright, C. I., & Barrett, L.
F. (2011). Differential hemodynamic response in affective circuitry with aging: An FMRI study
of novelty, valence, and arousal. Journal of Cognitive Neuroscience, 23(5), 1027-41.
doi:10.1162/jocn.2010.21527
Murphy, S. T., & Zajonc, R. B. (1993). Affect, cognition, and awareness: Affective priming with
optimal and suboptimal stimulus exposures. Journal of Personality and Social Psychology,
64(5), 723-39.
Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless
spreading activation and limited-capacity attention. Journal of Experimental Psychology:
General, 106(3), 226.
Norman. (2003). Emotional design: Why we love (or hate) everyday things. In Basic books; 1
edition.
Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Harvesting implicit group attitudes and
beliefs from a demonstration web site. Group Dynamics, 6, 101-115. doi:10.1037//1089-
2699.6.1.101
Öhman, A., Lundqvist, D., & Esteves, F. (2001). The face in the crowd revisited: A threat
advantage with schematic stimuli. Journal of Personality and Social Psychology, 80(3), 381.
doi:10.1037//0022-3514.80.3.381
Palmer, S. E., & Schloss, K. B. (2010). An ecological valence theory of human color preference.
Proceedings of the National Academy of Sciences of the United States of America, 107(19),
8877-82. doi:10.1073/pnas.0906172107
98
Patterson, K., Nestor, P. J., & Rogers, T. T. (2007). Where do you know what you know? The
representation of semantic knowledge in the human brain. Nature Reviews. Neuroscience,
8(12), 976-87. doi:10.1038/nrn2277
Pelli, D. G. (1997). The videotoolbox software for visual psychophysics: Transforming numbers
into movies. Spat Vis, 10(4), 437-42.
Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive control. In R. Solso (ed.),
Information processing and cognition: The loyola symposium. Hillsdale, NJ: Erlbaum. (pp.
55-85).
Ramachandran, V. S., & Hirstein, W. (1999). The science of art: A neurological theory of aesthetic
experience. Journal of Consciousness Studies, 6, 6(7), 15-51.
Rangel, A., Camerer, C., & Montague, P. R. (2008). A framework for studying the neurobiology of
value-based decision making. Nature Reviews. Neuroscience, 9(7), 545-56.
doi:10.1038/nrn2357
Rentschler, I., Jüttner, M., Unzicker, A., & Landis, T. (1999). Innate and learned components of
human visual preference. Curr Biol, 9(13), 665-71.
Rolls, E. T., & Grabenhorst, F. (2008). The orbitofrontal cortex and beyond: From affect to
decision-making. Prog Neurobiol, 86(3), 216-44. doi:10.1016/j.pneurobio.2008.09.001
Rudrauf, D., David, O., Lachaux, J. P., Kovach, C. K., Martinerie, J., Renault, B., & Damasio, A.
(2008). Rapid interactions between the ventral visual stream and emotion-related structures
rely on a two-pathway architecture. J Neurosci, 28(11), 2793-803.
doi:10.1523/JNEUROSCI.3476-07.2008
Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology,
39(6), 1161-1178.
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(5), 805-19.
99
Russell, J. A., & Carroll, J. M. (1999). On the bipolarity of positive and negative affect.
Psychological Bulletin, 125(1), 3-30.
Saito, Y. (2007). Everyday aesthetics. Oxford ; New York: Oxford University Press.
doi:9780199278350
Schama, S. (2010, January 22). What objects say about our times. Financial Times,
Sharot, T., De Martino, B., & Dolan, R. J. (2009). How choice reveals and shapes expected
hedonic outcome. J Neurosci, 29(12), 3760-5. doi:10.1523/JNEUROSCI.4972-08.2009
Sollberge, B., Rebe, R., & Eckstein, D. (2003). Musical chords as affective priming context in a
word-evaluation task. Music Perception, 20(3), 263-282.
Stenberg, G., Wiking, S., & Dahl, M. (1998). Judging words at face value: Interference in a word
processing task reveals automatic processing of affective facial expressions. Cognition and
Emotion, 12(6), 755ę782.
Sternberg, S. (1966). High-Speed scanning in human memory. Science, 153(3736), 652.
Sugrue, L. P., Corrado, G. S., & Newsome, W. T. (2005). Choosing the greater of two goods:
Neural currencies for valuation and decision making. Nature Reviews. Neuroscience, 6(5),
363-75. doi:10.1038/nrn1666
Tanaka, J. W., & Taylor, M. (1991). Object categories and expertise: Is the basic level in the eye
of the beholder? Cognitive Psychology, 23, 457-482.
Todorov, A., Said, C. P., Engell, A. D., & Oosterhof, N. N. (2008). Understanding evaluation of
faces on social dimensions. Trends in Cognitive Sciences. doi:10.1016/j.tics.2008.10.001
Todorov, Baron, & Oosterhof. (1985). Severe disturbance of higher cognition after bilateral frontal
lobe ablation: Patient EVR. Neurology, 35, 1731-1741. doi:10.1093/scan/nsn009
Unkelbach, C., Fiedler, K., Bayer, M., Stegmüller, M., & Danner, D. (2008). Why positive
information is processed faster: The density hypothesis. Journal of Personality and Social
Psychology, 95(1), 36-49. doi:10.1037/0022-3514.95.1.36
100
Unkelbach, C., von Hippel, W., Forgas, J. P., Robinson, M. D., Shakarchi, R. J., & Hawkins, C.
(2010). Good things come easy: Subjective exposure frequency and the faster processing of
positive information. Social Cognition, 28(4), 538-555.
Vanderwart, M. (1984). Priming by pictures in lexical decision. Journal of Verbal Learning and
Verbal Behavior, 23(1), 67-83.
Venables, P. H., & Christie, M. J. (1973). Mechanisms, instrumentation, recording techniques
and quantification of responses. In W. F. Prokasy & D. C. Raskin (Eds.), Electrodermal
activity in psychological research. (pp. 1-147). Academic Press.
Vessel, E. A., & Rubin, N. (2010). Beauty and the beholder: Highly individual taste for abstract,
but not real-world images. Journal of Vision, 10(2), 18.1-14. doi:10.1167/10.2.18
von Kriegstein, K., Kleinschmidt, A., Sterzer, P., & Giraud, A. L. (2005). Interaction of face and
voice areas during speaker recognition. Journal of Cognitive Neuroscience, 17(3), 367-76.
Weierich, M. R., Wright, C. I., Negreira, A., Dickerson, B. C., & Barrett, L. F. (2010). Novelty as a
dimension in the affective brain. Neuroimage, 49(3), 2871-8.
doi:10.1016/j.neuroimage.2009.09.047
Wentura, D. (2000). Dissociative affective and associative priming effects in the lexical decision
task: Yes versus no responses to word targets reveal evaluative judgment tendencies. J Exp
Psychol: Learn Mem Cogn, 26(2), 456-69.
Werner, J. S., & Chalupa, L. M. (2004). The visual neurosciences. Cambridge, Mass.: MIT Press.
Zajonc, R. B., & Markus, H. (1982). Affective and cognitive factors in preferences. Journal of
Consumer Research, 9(2), 123-131.
101
Different places the thesis was written.
Appendix 1. WordsAppendix 1. WordsAppendix 1. Words
Word Valence
Positive
paradise 8.72loved 8.64joy 8.6laughter 8.45champion 8.44mother 8.39cash 8.37orgasm 8.32romantic 8.32success 8.29pleasure 8.28treasure 8.27delight 8.26joyful 8.22graduate 8.19vacation 8.16rainbow 8.14joke 8.1cheer 8.1sex 8.05beach 8.03sexy 8.02hug 8engaged 8ecstasy 7.98aroused 7.97satisfied 7.94diamond 7.92snuggle 7.92merry 7.9party 7.86sunrise 7.86birthday 7.84beauty 7.82kindness 7.82adorable 7.81enjoyment 7.8spring 7.76sunlight 7.76car 7.73peace 7.72riches 7.7honor 7.66nature 7.65improve 7.65respect 7.64leader 7.63profit 7.63adventure 7.6beautiful 7.6kind 7.59money 7.59dog 7.57puppy 7.56sun 7.55holiday 7.55reward 7.53bed 7.51luscious 7.5memories 7.48good 7.47outdoors 7.47erotic 7.43elegant 7.43jolly 7.41silly 7.41couple 7.41warmth 7.41sky 7.37bride 7.34
Neutraljug 5.24arm 5.34cow 5.57owl 5.8door 5.13item 5.26chin 5.29fork 5.29time 5.31body 5.55rock 5.56frog 5.71book 5.72hand 5.95milk 5.95wine 5.95coin 6.02fish 6.04doll 6.09elbow 5.12paper 5.2table 5.22ankle 5.27truck 5.47paint 5.62hotel 6nurse 6.08grass 6.12locker 5.19doctor 5.2kettle 5.22pencil 5.22street 5.22theory 5.3poster 5.34violin 5.43basket 5.45avenue 5.5writer 5.52museum 5.54detail 5.55method 5.56yellow 5.61circle 5.67runner 5.67limber 5.68moment 5.76custom 5.85humble 5.86poetry 5.86window 5.91invest 5.93tennis 6.02bottle 6.15journal 5.14patient 5.29ketchup 5.6teacher 5.68nursery 5.73prairie 5.75trumpet 5.75whistle 5.81privacy 5.88garment 6.07curious 6.08umbrella 5.16material 5.26building 5.29bathroom 5.55mischief 5.57
Negativesad 1.61war 2.08fat 2.28rape 1.25sick 1.9dead 1.94jail 1.95bomb 2.1pain 2.13debt 2.22hell 2.24lice 2.31rage 2.41ugly 2.43ache 2.46rude 2.5death 1.61slave 1.84drown 1.92upset 2vomit 2.06toxic 2.1thief 2.13devil 2.21whore 2.3crash 2.31anger 2.34roach 2.35alone 2.41agony 2.43cancer 1.5misery 1.93poison 1.98afraid 2burial 2.05prison 2.05trauma 2.1corpse 2.18victim 2.18rotten 2.26stupid 2.31robber 2.61suicide 1.25funeral 1.39torture 1.56poverty 1.67failure 1.7tragedy 1.78seasick 2.05hostage 2.2crushed 2.21divorce 2.22traitor 2.22violent 2.29enraged 2.46illness 2.48tornado 2.55rejected 1.5mutilate 1.82bankrupt 2headache 2.02helpless 2.2sickness 2.25dreadful 2.26massacre 2.28deserter 2.45hardship 2.45burdened 2.5addicted 2.51jealousy 2.51
Appendix 2. ObjectsAppendix 2. ObjectsAppendix 2. Objects
Image Valence
Positive
0.74
0.71
0.73
0.79
0.8
0.78
0.77
0.71
0.72
0.76
0.72
0.75
0.74
0.71
0.73
0.77
0.75
0.78
0.79
0.72
0.78
0.73
0.72
0.74
0.77
0.77
0.76
0.82
0.79
0.75
0.77
0.75
0.74
0.76
0.75
0.76
0.79
0.73
0.71
0.75
0.72
0.71
0.74
0.75
0.74
0.75
0.72
0.74
0.73
0.8
0.74
0.71
0.8
0.76
0.78
0.72
0.82
0.78
0.77
0.79
0.68
0.7
0.69
0.69
0.7
0.69
0.69
0.64
0.71
0.76
0.7
0.72
0.63
0.75
0.68
0.77
0.64
0.7
0.7
0.73
0.67
0.67
0.72
0.64
0.7
0.67
0.72
0.66
0.63
0.66
0.68
0.68
0.69
0.63
0.75
0.73
0.67
0.73
0.72
0.71
0.64
0.7
0.73
0.75
0.7
Negative
0.11
0.19
0.16
0.22
0.23
0.23
0.12
0.13
0.12
0.21
0.26
0.22
0.18
0.25
0.15
0.13
0.15
0.14
0.17
0.15
0.18
0.24
0.24
0.12
0.17
0.23
0.16
0.25
0.19
0.19
0.2
0.19
0.26
0.14
0.17
0.25
0.14
0.1
0.21
0.12
0.13
###
0.17
0.15
0.18
0.14
0.17
0.23
0.25
0.22
0.23
0.16
0.17
0.18
0.13
0.17
0.18
0.12
0.07
0.17
0.3
0.29
0.29
0.28
0.27
0.29
0.26
0.27
0.29
0.32
0.32
0.28
0.27
0.27
0.32
0.28
0.2
0.25
0.22
0.29
0.3
0.25
0.22
0.26
0.25
0.26
0.29
0.18
0.24
0.21
0.25
0.2
0.24
0.2
0.21
0.23
0.21
0.19
0.19
0.19
0.2
0.26
0.19
0.2
0.26
Appendix 3. Micro-valence Objects
Image Valence
Positive
2.30
2.14
2.08
1.95
1.76
1.68
1.65
1.54
1.49
1.43
1.41
1.35
1.32
1.32
1.30
1.27
1.27
1.24
1.16
1.11
1.08
1.00
0.92
0.92
0.92
0.92
0.81
0.81
0.73
0.68
0.57
0.51
0.49
0.46
0.43
0.43
0.43
0.41
0.35
0.32
0.32
0.30
0.27
0.27
0.24
0.24
0.22
0.22
0.19
0.16
0.16
0.16
0.16
0.11
0.11
0.11
0.00
Negative
-2.54
-2.38
-2.27
-2.11
-1.95
-1.81
-1.70
-1.68
-1.54
-1.38
-1.35
-1.27
-1.24
-1.24
-1.22
-1.22
-1.19
-0.92
-0.92
-0.89
-0.89
-0.86
-0.86
-0.84
-0.81
-0.76
-0.70
-0.65
-0.62
-0.59
-0.57
-0.49
-0.49
-0.46
-0.46
-0.43
-0.43
-0.43
-0.43
-0.43
-0.41
-0.38
-0.38
-0.38
-0.35
-0.35
-0.35
-0.30
-0.22
-0.22
-0.19
-0.19
-0.16
-0.16
-0.14
-0.14
-0.11
-0.08
-0.05
-0.05
-0.05
-0.05
-0.03