Graduate Theses and Dissertations
Iowa State University Capstones, Theses and
Dissertations
2019
The non-specificity of prosopagnosia: Can
prosopagnosics distinguish sheep?
Alexander Robert Toftness
Iowa State University
Follow this and additional works at: https://lib.dr.iastate.edu/etd
Part of the Cognitive Psychology Commons, and the Neuroscience and Neurobiology Commons
This Thesis is brought to you for free and open access by the Iowa State University Capstones, Theses and Dissertations at Iowa State University Digital
Repository. It has been accepted for inclusion in Graduate Theses and Dissertations by an authorized administrator of Iowa State University Digital
Repository. For more information, please contact digirep@iastate.edu.
Recommended Citation
Toftness, Alexander Robert, “The non-specificity of prosopagnosia: Can prosopagnosics distinguish sheep?” (2019). Graduate Theses
and Dissertations. 17109.
https://lib.dr.iastate.edu/etd/17109
The non-specificity of prosopagnosia: Can prosopagnosics distinguish sheep?
by
Alexander R. Toftness
A thesis submitted to the graduate faculty
in partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE
Major: Psychology
Program of Study Committee:
Eric E. Cooper, Major Professor
Christian A. Meissner
Jonathan W. Kelly
The student author, whose presentation of the scholarship herein was approved by the
program of study committee, is solely responsible for the content of this thesis. The
Graduate College will ensure this thesis is globally accessible and will not permit
alterations after a degree is conferred.
Iowa State University
Ames, Iowa
2019
Copyright © Alexander R. Toftness, 2019. All rights reserved.
ii
TABLE OF CONTENTS
Page
LIST OF FIGURES ……………………………………………………………………………………………… iii
LIST OF TABLES
………………………………………………………………………………………………..
iv
ACKNOWLEDGMENTS ………………………………………………………………………………………v
ABSTRACT…………………………………………………………………………………………………………
vi
CHAPTER 1. INTRODUCTION …………………………………………………………………………… 1
The Coordinate Relations Hypothesis ……………………………………………………………….. 3
Alternatives to the Coordinate Relations Hypothesis …………………………………………… 7
Prosopagnosic Experiments ……………………………………………………………………………. 14
The Present Experiments ……………………………………………………………………………….. 17
CHAPTER 2. EXPERIMENT 1 …………………………………………………………………………… 20
Method ………………………………………………………………………………………………………… 20
Participants……………………………………………………………………………………………… 20
Test of prosopagnosia …………………………………………………………………………. 20
Nature of the acquired prosopagnosia……………………………………………………. 20
Age-matched and gender-matched control …………………………………………….. 21
Materials…………………………………………………………………………………………………. 21
Design and Procedure ………………………………………………………………………………. 24
Results
…………………………………………………………………………………………………………. 24
Discussion ……………………………………………………………………………………………………. 30
CHAPTER 3. EXPERIMENT 2 …………………………………………………………………………… 32
Method ………………………………………………………………………………………………………… 32
Participants……………………………………………………………………………………………… 32
Materials…………………………………………………………………………………………………. 32
Design and Procedure ………………………………………………………………………………. 35
Results
…………………………………………………………………………………………………………. 35
Discussion ……………………………………………………………………………………………………. 43
CHAPTER 4. GENERAL DISCUSSION ……………………………………………………………… 46
REFERENCES ………………………………………………………………………………………………….. 50
APPENDIX. INSTITUTIONAL REVIEW BOARD APPROVAL…………………………… 57
iii
LIST OF FIGURES
Page
Figure 1
The Coordinate Relations Hypothesis ……………………………………………….. 5
Figure 2
Reference Points in the Coordinate Relations Hypothesis ……………………. 6
Figure 3
Experiment 1 Legislator Face Stimuli ……………………………………………… 23
Figure 4
Experiment 1 Overall Results …………………………………………………………. 26
Figure 5
Experiment 1 Trial Level Results ……………………………………………………. 27
Figure 6
Experiment 2 Sheep Face Stimuli
……………………………………………………. 33
Figure 7
Experiment 2 Sheep Bodies Exemplars
……………………………………………. 34
Figure 8
Experiment 2 Overall Results …………………………………………………………. 37
Figure 9
Experiment 2 Trial Level Results ……………………………………………………. 38
Figure 10
Experiment 2 Trials by Body Parts Results ………………………………………. 39
iv
LIST OF TABLES
Page
Table 1 Accuracy in Experiment 1 by Trial Type and Correct Response ………………… 28
Table 2 Accuracy in Experiment 2 by Trial Type and Correct Response ………………… 40
v
ACKNOWLEDGMENTS
I would like to thank my committee chair, Dr. Eric Cooper, and my committee members,
Dr. Chris Meissner, and Dr. Jon Kelly.
I would like to thank the Iowa State Sheep Teaching Farm for allowing the
photographing of their animals during materials creation. I would also like to thank Charles
“Joey” Peasley and Dr. Eric Cooper for their roles in the creation of the sheep photo stimuli.
Thank you to the age-matched control who agreed to participate in this project. Many thanks to
my partner, family, and friends for their support over the years.
Finally, a special thank you to LB, the prosopagnosic participant in these experiments,
without whom this project would not have been possible.
vi
ABSTRACT
The impact that prosopagnosia (face-blindness) has on the human visual system has long
been hypothesized with regard to the specifics of the impairment. The leading hypothesis in the
literature, the face-specificity hypothesis, proposes that prosopagnosia is specific only to human
faces. Other hypotheses have offered alternative explanations for what sorts of identification
tasks might be affected by damage to the fusiform face area resulting in prosopagnosia, including
the biological recognition, expert recognition, and subordinate-level recognition hypotheses. An
additional hypothesis, the coordinate relations hypothesis, offers a compelling explanation for
the underlying process disrupted by prosopagnosia: that the brain’s ability to detect metric
changes has been damaged resulting in a deficit to face recognition. This hypothesis was tested
by looking for deficits in performance in a prosopagnosic when identifying non-faces, because
such differences would not be explained by the face-specificity hypothesis. Therefore, sheep
faces were used as a class of stimuli to explore whether prosopagnosia affects identification of
sheep faces in much the same way that it affects identification of human faces. Results of two
experiments showed that a prosopagnosic was impaired on identification of sheep faces,
providing support for the coordinate relations hypothesis.
1
CHAPTER 1. INTRODUCTION
As we continue to learn about the hardware of the brain that makes cognition possible,
the non-specificity of brain functions is increasingly revealed. While brain functions can
certainly be localized to an extent, the brain is not neatly divided into modular sections that
specialize in one function and one function only. We are now in an age of neuroplasticity and the
connectome where it is increasingly difficult to label any part of the brain with a specific
function because of the interconnectedness of processes. Instead, complex networks, varying not
only between individual humans, but also across the entire lifespan of those humans, are now
implicated in multiple tasks as opposed to one specific task. A veritable bevy of imaging
techniques (e.g., fMRI, etc.) have revealed that the cerebellum, once known only for motor
control, is involved in a wide variety of neurological functions (Schmahmann, 2016), that the
amygdala, once known only as an emotional center, is important for memory, learning, pain,
motivation, and other deeply interconnected processes (Amunts et al., 2005; Richardson, 1973),
and even the very concept of morality, once thought to be housed in some vessel of the soul, is
not a localized area whatsoever, but arises from multiple processes in distributed locations
(Greene, 2015). But there are a few regions of the brain that have dedicated advocates insisting
upon specific assigned functions. Chiefly among these is the fusiform face area (FFA) and its
neighboring regions implicated in the processing of human faces (e.g., Schalk et al., 2017;
Kanwisher, 2017, Kanwisher, McDermott, & Chun, 1997).
One reason that some consider the FFA to be specific to human face processing is that the
brain process responsible for the recognition of faces has been shown to differ from the process
used to recognize other objects. Cases of prosopagnosia (face-blindness) and object agnosia
2
(object-blindness) show a double dissociation suggesting that the brain uses separate processes
(or recognition subsystems) to accomplish face and object recognition (Farah, 2004).
Behavioral evidence for separate recognition systems also exists. Inverting (Yin, 1969) or
creating photographic negatives (Galper & Hochberg, 1971) disrupt face recognition more than
object recognition. If part of a face such as a nose is learned in the context of a face, testing
recognition of the nose in isolation produces a recognition deficit when compared to testing
recognition in the context of the whole face suggesting encoding of the relationships between
parts beyond the features of the nose itself (i.e., the part-whole effect, Tanaka & Farah, 1993).
Identifying half-faces is slower when the tested half-face is aligned with another half-face
(creating a whole face) than if the halves are misaligned (i.e., the composite effect, Young,
Hellawell, & Hay, 1987). Visual half-field studies have found that presenting faces to the left
visual field (and thus to the right hemisphere) as opposed to the right visual field leads to better
recognition performance, but no hemisphere advantage is typically found for basic-level object
recognition (Brooks & Cooper, 2006).
Physiological evidence for this dissociation between face and object recognition comes
from a variety of methods that have isolated brain responses to faces in the FFA and nearby brain
regions. The FFA shows maximal activation to faces when compared to non-face objects (e.g.,
flowers, houses, chairs, etc.) in studies using fMRI (McCarthy, Puce, Gore, & Allison, 1997;
Pinsk et al., 2009), magnetoencephalography (Uono, Sato, Kochiyama, Kubota, Sawada,
Yoshimura, & Toichi, 2017), single neuron recording (in monkeys: Földiák, Xiao, Keysers,
Edwards, & Perrett, 2004), and ERP methodologies (Sadeh, Podlipsky, Zhdanov, & Yovel,
2010). Indeed, intracranial electroencephalography (iEEG) has shown a wide distribution of
neurons that respond to faces over other categories of objects throughout the ventral
3
occipitotemporal cortex (Rossion, Jacques, & Jonas, 2018), and electrical stimulation of these
areas can produce temporary hallucinations of faces superimposed on viewed objects (Schalk et
al., 2017). It is no wonder, with the physiological evidence pointing firmly in favor of face-
specificity, that the FFA has been considered by some researchers to be an area devoted
exclusively to face identification.
We know that these processes for object recognition and face recognition are different.
However, the issue with which this proposal is concerned is whether the underlying process
responsible for the successful identification of faces is, in fact, unique to human faces. One
theory that can explain how these two recognition systems differ is the coordinate relations
hypothesis, which is not specific to faces.
The Coordinate Relations Hypothesis
The coordinate relations hypothesis states that there are two recognition systems that the
brain can use to accomplish recognition tasks, and the system that the brain uses to complete a
task is determined by the computational demands of the task in question. The two recognition
systems used to explain the dissociation between object identification and face identification are
the categorical system and the coordinate system. These two systems were initially proposed by
Kosslyn (1987) as lateralized (hemisphere dominant/specialized) systems used for both seeing
and imagining (i.e., spatial relations theory). According to the coordinate relations hypothesis, if
a visual recognition task can be accomplished using a representation of an object’s parts and the
categorical relations of those parts, then the brain will use a basic-level recognition system that
does not represent exact distances. This first system is the categorical system, which encodes
visual primitives (e.g., geons, Biederman, 1987) and the relations between those primitives (e.g.,
above, below, side of). However, if the task requires distinguishing among objects that have the
4
same parts and relations (see Figure 1), then the face identification system will be used. This
second system is the coordinate system, which encodes metric data from the visual input, such as
how far primitives in the representation are from a reference point (Brooks & Cooper, 2006;
Cooper & Brooks, 2004; Cooper & Wojan, 2000). The basic-level system is more efficient, and
so is used to satisfice a visual task whenever possible. However, the coordinate-relations
hypothesis predicts that there is nothing special per se about human faces, but that rather, faces
are generally subjected to the use of the coordinate system because the task of facial recognition
demands it (See Figure 2).
5
Figure 1. Telling a person apart from a sheep is a trivial matter because of categorical
differences in structure (e.g., face shape, etc.), but telling a person apart from another person or a
sheep apart from another sheep requires differentiating between the same parts with the same
relations (e.g., two eyes above a nose above a mouth). Accomplishing this task requires
additional information, namely the metric distances between the parts (e.g., the amount of space
between the eyes).
6
Figure 2. The coordinate relations hypothesis posits that metric distances in visual information
can be encoded. For example, for the left-side woman, the distance from the left eye to the right
eye is 1.5 grid units. For the right-side woman, the distance from the left eye to the right eye is 1
grid unit. This difference could be used to differentiate these women. Similarly, the coordinate
relations hypothesis would predict that metric differences in sheep faces could be encoded and
used to distinguish the sheep faces.
7
Alternatives to the Coordinate Relations Hypothesis
Previously proposed hypotheses concerning the specificity of the FFA include the
biological recognition hypothesis (Farah, McMullen, & Meyer, 1991), the expert recognition
hypothesis (Diamond & Carey, 1986; Gauthier & Tarr, 1997), the subordinate-level recognition
hypothesis (Gauthier, Anderson, Tarr, Skudlarski, & Gore, 1997; Gauthier, Tarr et al., 2000), and
the face-specificity hypothesis (Kanwisher et al., 1997; Xu, Liu, & Kanwisher, 2005). Each of
these hypotheses, and the general pattern of findings supporting each, are discussed in turn.
The biological recognition hypothesis proposes that it is not faces, per se, that the FFA
responds to, but rather, biological stimuli in general such as plants and animals as opposed to
non-living things (Farah, Meyer, & McMullen, 1996). Some evidence for this theory is that
prosopagnosics, having lesions of the FFA, often have difficulty differentiating between animals
(Farah et al., 1991; Pallis, 1955), especially four-legged animals (Chao, Martin, & Haxby, 1999).
Some prosopagnosics also have difficulty differentiating among fruits and vegetables (Barton,
2008). While there are regions of the brain that seem to react preferentially to biological stimuli,
such as biological movement (Grossman et al., 2000), and studies have found deficits that seem
to affect naming of biological stimuli more than naming of non-living stimuli in the form of
category-specific visual agnosia or anomia (Caramazza & Shelton, 1998; Farah et al., 1991;
Wolk, Coslett, & Glosser, 2005), the FFA does not appear to be an area specializing in biological
stimuli. In terms of specificity, the FFA responds more strongly to faces than to other biological
stimuli, such as hands (Kanwisher et al., 1997). Also, prosopagnosics can differentiate between
certain biological stimuli more readily than others as a function of the featural differences
between them (e.g., differentiating a two-legged animal from a four-legged animal is easier for a
prosopagnosic than differentiating between two four-legged animals, Casner, 2006). In terms of
8
non-biological stimuli, prosopagnosics can also show deficits in entirely non-living categories
such as cars (de Haan & Campbell, 1991), buildings, and money (O’Brien, 2018), and other
visually ambiguous stimuli (Damasio, Damasio, & Van Hoesen, 1982), demonstrating that
damage to the FFA does not specifically target the recognition of biological stimuli. Therefore, a
hypothesis restricting the FFA’s role to recognition of biological stimuli is not sufficient to
explain the findings in the literature. The coordinate relations hypothesis can explain the finding
that prosopagnosics have difficulty distinguishing between living things (people, four-legged
animals, fruits) as a consequence of the fact that many living things share the same parts and
relations (e.g., people generally have two eyes, a nose, and a mouth in a specific configuration;
four legged animals have four legs attached to the same points on the body; an apple and a
tomato may both be represented as a stem above a spherical body) and therefore rely on the
damaged (in prosopagnosia) coordinate-based recognition system for effective differentiation.
The expert recognition hypothesis (also called the expertise hypothesis, e.g., Xu et al,.
2005) says that the FFA is responsible for recognition of classes of objects that people are
experts at identifying, and that faces are a class of objects that most humans are experts at
recognizing (Gauthier, 2017; Gauthier & Tarr, 1997). Evidence for the expert recognition
hypothesis comes from two groups of studies. The first involves testing experts in categories of
stimuli (such as dog experts, car experts, or bird experts) using paradigms that elicit processing
deficits for faces, typically inversion effects, in order to determine if a similar deficit will show
up for non-face classes of stimuli (e.g., Diamond & Carey, 1986; McGugin, Newton, Gore, &
Gauthier, 2014). The second group of studies involve expertise training for a class of objects,
often an artificial class (e.g., using nonsense objects made of geons that can be easily
interchanged to create novel stimuli), followed by testing of sensitivity for recognition of highly
9
similar objects within that class (e.g., “greeble” studies, Gauthier & Tarr, 1997; Tarr & Gauthier,
2000). Both of these groups of studies have found activation in the FFA, to some degree, in the
absence of faces.
The subordinate-level recognition hypothesis proposes that the recognition system
responsible for distinguishing individual facial identities is not used for recognizing faces, per se,
but that facial identity happens to be one category of stimuli requiring an extra level of
processing, referred to as subordinate-level processing, beyond basic-level processing used for
most object recognition tasks (Gauthier et al., 1997). According to this account, subordinate-
level recognition occurs in the FFA region impacted by prosopagnosia. In other words, while the
basic-level system was said to be responsible for distinguishing amongst members of disparate
classes (a chair among tables, for example), the subordinate-level system was hypothesized to be
responsible for distinguishing within classes of items sharing a category (a rocking chair
amongst reclining chairs, for example) (Gauthier, Tarr et al., 2000). In this way, the tasks of
detecting a face (a basic-level system task) and identifying a face (a subordinate-level task) could
be differentiated from each other, potentially explaining why prosopagnosia seems to impact
tasks wherein the differentiation is between minute differences (e.g., between cars of the same
model from different years, Gauthier, Skudlarski, Gore, & Anderson, 2000; etc.). In support of
this proposal, research has shown that face detection and face identification are indeed
dissociable (Robertson, Jenkins, & Burton, 2017).
These two hypotheses of expertise and subordinate-level recognition are somewhat
compatible with one another, and Tarr and Gauthier (2000) made the case that the face
recognition area of the brain (the FFA) was actually a flexible subordinate-level visual
processing center that was mediated by expertise, and this could include faces as well as other
10
classes of stimuli. In other words, their merged hypothesis was that the development of expertise
in recognizing a class of objects (e.g., differentiating between families of greebles, or between
bird species, or any class of objects) was reflected in the brain’s FFA by training the subordinate-
level visual processing system to differentiate amongst the minute differences between objects
sharing a class. This process, in turn, was said to also explain face-processing, simply as one
class of expert objects that the average person had expertise in due to training over their lives
(i.e., extensive exposure to faces and need to differentiate them, leading to expertise).
Certain aspects of the expertise and subordinate-level hypotheses are supported by the
empirical data. Some fMRI studies have shown that manipulating levels of expertise produce
more successful subordinate-level categorizations with corresponding increases in the response
of the FFA (Tarr & Gauthier, 2000; Xu, 2005). Level of expertise in cars has been shown to have
a significant predictive relationship with the strength of brain response in the FFA (McGugin et
al., 2014). Greeble studies have shown that training on a novel class of stimuli created sensitivity
to inversion effects and small deviations in their configurations, similar to faces (Gauthier &
Tarr, 1997).
However, evidence against the expertise and subordinate-level hypotheses is also
extensive. Some neuroimaging studies have failed to find that brain activity in the FFA, in
response to a class of objects, increases as a function of level of expertise with that class of
objects (Grill-Spector, Knouf, & Kanwisher, 2004), and inversion effects do not always correlate
with expertise (Rezlescu, Chapman, Susilo, Caramazza, 2016; Weiss, Mardo, & Avidan, 2016).
Greeble studies have been criticized because the greeble stimuli resemble faces, which could be
why they recruit activity in the FFA (Kanwisher & Yovel, 2006; but see Gauthier, Behrmann, &
Tarr, 2004), and because Greeble expertise has been acquired successfully by a prosopagnosic
11
with an impaired FFA (Duchaine, Dingle, Nakayama, & Butterworth, 2004). Deficits resulting
from prosopagnosia, too, seem to exist regardless of expertise levels and subordinate classes. For
example, deficits have been shown to exist for prosopagnosics when expertise and class remain
constant, via metric manipulations of within-class objects such as furniture (e.g., changing the
length of a table) that create a deficit while structural changes (e.g., changing the shape of the
table) do not produce this deficit (Casner, 2006). Additionally, prosopagnosics have shown
deficits in basic level naming tasks of animals, such as between species of four-legged animals
(Chao et al., 1999), and in situations where the subordinate-level recognition hypothesis does not
predict differences, because the animal discrimination is at the basic level (e.g., fox vs. dog,
Casner, 2006). Therefore, while the FFA and closely related regions of the brain appear to use
expertise during identification, and subordinate-level familiarity is necessary for effective
identification (e.g., semantic naming tasks benefit from it), they do not appear sufficient to
explain prosopagnosia-related deficits to face identification.
The coordinate relations hypothesis and subordinate-level recognition hypothesis both
predict that the face recognition regions will be used to make subordinate level distinctions
among objects sharing the same parts and relations. In general, both predict the empirical finding
that subordinate-level recognition tasks (e.g., tasks that require the use of metric differences)
show more activation in the FFA than do basic-level recognition tasks (e.g., Gauthier et al.,
1997). However, the coordinate relations hypothesis differs in its prediction when the
subordinate-level recognition task can be satisficed by non-metric features (i.e., categorical
features). For example, the coordinate relations hypothesis predicts that subordinate-level
distinction between objects that do not share the same parts and categorical relations should
12
instead be performed by the basic level object (i.e. categorical) recognition system, which
satisfices the recognition task with less computational demand.
The coordinate relations hypothesis agrees with the expertise recognition hypothesis that
prosopagnosics’ deficits are not specific to human faces (i.e., differentiation of non-faces can be
affected). However, the coordinate relations hypothesis does not predict that expertise is a
necessary or sufficient condition for a prosopagnosic deficit to occur. While expertise does
physically modify the brain (e.g., by increasing the complexity of neurons responding to a task as
expertise in that task increases, McGugin et al., 2014), and could therefore change how the brain
responds when identifying a stimulus (e.g., by increasing reliance on neurons localized in the
FFA), the coordinate relations hypothesis predicts that a deficit to identification can occur in the
absence of expertise. Importantly, the differences used for differentiation in expert recognition
tasks commonly used in expertise studies (e.g., greebles, dogs, birds, cars) can frequently be
distinguished by coordinate, but not categorical, relations, allowing the coordinate relations
hypothesis to potentially explain the involvement of the FFA (Brooks & Cooper, 2006).
The face-specificity hypothesis has a very straightforward account of the FFA’s function.
Essentially, the face-specificity hypothesis posits that the FFA contains a mechanism for
detecting the specific ‘holistic’ geometry of faces, and that this mechanism exclusively functions
for human faces (Kanwisher & Yovel, 2006). Additionally, what exactly is meant by the word
‘holistic’ (as well as the word configural) is controversial, in that the operational definition varies
greatly between studies (Esins, Schultz, Stemper, Kennerknecht, & Bülthoff, 2016; O’Toole,
Abdi, Deffenbacher, & Valentin, 1995). Because the word holistic means different things to
different researchers, and has been used to refer to both theoretical constructs as well as
measurements, it is not a useful term for creating predictions related to what is physically
13
different about the face that can be put to an empirical test (see Richler, Palmeri, & Gauthier,
2012, for a discussion of the “loosely defined” definition of holistic processing). For that reason,
the word holistic has been specifically avoided in the preparation of these experiments, in favor
of testable operational definitions (e.g., categories of up, down; metric distance) without
terminology baggage.
There is much evidence against the face-specificity account. The FFA contains cells that
are not face selective (Weiner & Grill-Spector, 2010). Prosopagnosics show recognition deficits
in many areas that are not related to faces, such as birds (Bornstein, 1963), cows (Bornstein,
Sroka & Munitz, 1969) cars (Damasio et al., 1982), foods when grocery shopping (Damasio et
al., 1982), chairs (Faust, 1955), and buildings (Pallis, 1955). Inversion effects can show up when
using categories of stimuli that are not human faces, such as dogs (Diamond & Carey, 1986) and
cars (Rezlescu et al., 2016). Prosopagnosics make more errors than neurotypical people in
detecting differences when parts of non-face objects change position or orientation (e.g., metric
changes such as an oblique angle of one geon relative to another geon becoming larger such that
it is still an oblique angle) but do not make more errors when the changes are categorical (e.g., an
oblique angle of one geon relative to another geon becoming perpendicular, Casner, 2006). The
face specificity hypothesis cannot account for these findings. Furthermore, there is no definitive
reason that a region of the brain typically used to process faces cannot also sometimes be called
upon for the processing of other stimuli (Bruce & Humphreys, 1994). In other words,
experimentally showing that faces are processed in the FFA does not, and cannot, “prove” that
other categories of objects are not processed there.
Whenever a claim is made that a prosopagnosic has a deficit specific to human faces and
only human faces, it can always be asked whether they would have shown a deficit with another
14
category of objects that they were not tested upon (Barton & Corrow, 2016). Interestingly, a
meta-analysis including 72 people with face recognition deficits that were also tested in their
recognition of at least three other object categories revealed that zero of the people showed a
pure face recognition deficit without also having some other object recognition deficit (Geskin &
Behrmann, 2018). In theory, a single “pure” face recognition deficit due to FFA damage would
confirm the face specificity hypothesis, but this cannot be accomplished due to the impossibility
of testing people with face recognition deficits on all possible recognition tasks. Therefore, the
face specificity hypothesis necessarily has weak credibility in its status as a theory.
Importantly, studies that implicate face-specific mechanisms in prosopagnosia have not
tested the coordinate relations hypothesis’ prediction of impairment in the differentiation of
images with metric differences, even though they claim to have ruled out alternative hypotheses
(e.g., Duchaine, Yovel, Butterworth, & Nakayama, 2006; Kanwisher, 2000; Kanwisher & Yovel,
2006).
The coordinate relations hypothesis is not compatible with the face-specificity
hypothesis. According to the coordinate relations hypothesis, there is nothing special about faces
specifically that would lead to the use of the proposed coordinate system, but instead, any
differentiation task requiring attention to metric differences that cannot be satisficed via a less
computationally expensive difference (e.g., color, shape) will use the coordinate system.
In order to test these hypotheses and determine what the specific (or not-so-specific)
function of the FFA is, it is informative to use experiments with prosopagnosic participants.
Prosopagnosic Experiments
There are two main groups of prosopagnosics: acquired (i.e., from brain damage) and
congenital (i.e., developmental). Both show physiological differences from neurotypical brains in
15
the FFA. Acquired prosopagnosics show lesions in these brain regions implicated in face
recognition especially including the FFA of the right hemisphere (Barton, 2008) while congenital
prosopagnosics show reduced gray matter in the FFA and related brain regions (Garrido et al.,
2009) and impaired (undifferentiated) neural response patterns to intact and scrambled faces
(Avidan & Behrmann, 2009; Zhang, Liu, & Xu, 2015).
In prosopagnosics, there is a dissociation between face identification and face detection.
In other words, prosopagnosics can detect faces even when they cannot identify the faces (e.g.,
Garrido, Duchaine, & Nakayama, 2008). This dissociation between identification and detection
is supported via physiological and neuropsychological evidence. Physiological evidence includes
that transmagnetic stimulation of the OFA disrupts recognition (identification) of faces, but not
categorization (detection) of faces (Solomon-Harris, Mullin, & Steeves, 2013). Experimentally,
it has been shown that the ability to detect and the ability to identify faces are not significantly
related (Robertson, Jenkins, & Burton, 2017). Neuropsychological evidence comes from the fact
that while prosopagnosics struggle to identify faces, they can detect faces and facial expressions,
as well as individual facial features such as eyes and eye color (Richoz, Jack, Garrod, Schyns, &
Caldara, 2015; Tranel, Damasio, & Damasio, 1988).
One of the main challenges to the coordinate relations hypothesis is McNeil &
Warrington’s (1993) finding that a prosopagnosic was able to distinguish sheep. In this study, a
prosopagnosic, known as WJ, was tested for his ability to recognize faces of people and faces of
sheep. Upon comparing his results to a control sample (including some age-controlled and
profession-controlled participants) he was found to have a deficit in recognizing human faces,
but not sheep faces. Specifically, WJ was asked to complete a number of sheep identification
tasks. The tasks were completed using sheep photos without color, showing only the heads of the
16
sheep (J. E. McNeil, personal communication, February 5, 2019). In one experiment, WJ saw a
sequential presentation of 8 sheep photos, and then later saw a presentation of 16 sheep photos
and had to indicate whether each photo had been in the previous list. This task was repeated a
total of three times: using sheep from WJ’s own flock, using sheep unfamiliar to WJ, and using
human faces unfamiliar to WJ. WJ performed better with both sheep tasks than with the human
face task, while control subjects performed better with human faces and much worse with the
sheep faces (both compared to human face performance and compared to WJ’s sheep
performance). In a second experiment, WJ memorized names associated with sheep faces and
then was asked to recall those names when shown the faces later. This process was repeated with
human faces, as well. The results showed that control subjects outperformed him at the human
face task but scored similarly on the sheep task. The study concluded that prosopagnosia is a
disorder specific to human faces.
However, this study by McNeil & Warrington (1993), despite being called ingenious in
the prosopagnosia literature (Farah, 2004) and being cited over one hundred and fifty times at the
time of this writing, has several glaring flaws that could be remedied by an improved design.
First, the individual sheep used in the experiment were not represented by multiple pictures, but
were instead represented by a single exemplar photograph, and the photographs were not
standardized in terms of aspect ratio or background detail (i.e., features of the photo that were
not the sheep in question varied from photo to photo). This procedure is problematic because a
participant (prosopagnosic or not) could memorize certain non-face features of the photo in order
to improve their accuracy during the trials. Secondly, the prosopagnosic tested in this experiment
provided some of the sheep photos that were used for the study. This problem compounds with
the first problem, in that his familiarity with the photos could potentially boost his accuracy.
17
Thirdly, the sheep photos used had noticeable identifiable features, such as spots, that would
allow the possibility of recognition via systems other than the traditional face recognition system.
That is, people could memorize a single spot on the image, and use that spot to ascertain the
identity of the same individual later. Fourthly, given that only one picture was used for each
sheep, the orientation or pose of the sheep could be used to identify specific individuals.
Therefore, while the research question of McNeil and Warrington was intriguing, the way in
which they designed the experiment to answer the question was problematic.
The Present Experiments
The current research attempts to replicate McNeil & Warrington (1993), but with
superior materials including multiple photos with various viewpoints of each individual sheep to
be used in the experiment and control of exposure to those materials. This procedure can
determine whether prosopagnosia interferes with recognition of sheep faces in the absence of
other surface cues (e.g., spots).
A second improvement is to use a paradigm more traditionally suited to studies
concerning face identification. In the McNeil and Warrington (1993) paper, participants reported
whether they recognized a face from a recently displayed group of faces, or matched learned
names with previously seen faces. While it is certainly true that ecologically valid face
identification can take place as a comparison between a stored representation and new input from
the environment, this design depends on successful memory as well. Research has shown that
face memory (sequential presentation) and face perception (simultaneous presentation) tasks
correlate highly (r = .61 in Bowles et al., 2009), and both can be useful for the assessment of face
recognition impairments, but it may be argued that remembering details of a sheep face is
inherently harder than memorizing a human face, as a person may not know which details to
18
attend to. In order to eliminate the extra task of memorizing and recalling, instead of presenting
photos sequentially as in the McNeil and Warrington paper, it is more appropriate to use
simultaneous presentation of two side-by-side photographs as are used in both forensic casework
(e.g., legal contexts) and laboratory studies of human face recognition (White, Norell, Phillips &
O’Toole, 2017). It has been recommended that experiments designed specifically for
prosopagnosics should use identification tasks featuring multiple simultaneously presented
images showing different viewpoints of the subject (Watson, Huis in ‘t Veld, & de Gelder,
2016), eliminating both general memory problems (e.g., sheep being harder to remember,
inaccurate memory retrieval, etc.) and a memorization problem (i.e., participants will not be able
to memorize features of the photograph itself due to different viewpoints of the individuals being
used). Doing so will allow a direct test of whether identity can be ascertained from a set of two
images of the same (or different) individual(s), and whether the mechanism of doing so differs
depending on whether the image is of a human face or of a sheep face.
A human face identification task and a sheep face identification task were used to test
whether the original findings from the McNeil & Warrington (1993) paper replicate when
surface cues on the sheep cannot be used to perform the task. The human face experiment
occurred first, in order to determine if the paradigm used (simultaneous presentation for a
same/different identification task) produced the expected deficit in human face identification for
the prosopagnosic participant.
In order to test whether the inclusion of spots assists a prosopagnosic in determining the
identify of sheep, the sheep experiment included comparisons between sheep with spots and
sheep without spots. The spotted (speckled-faced) sheep and the non-spotted (white-faced) sheep
were compared within and between groups. That is, speckled-white, white-white, and speckled-
