quality dependent colored - university of toronto
TRANSCRIPT
This manuscript has been reproduced from the microfilm master. UMI films the
text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
cornputer printer.
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleedthrough, substandard margins, and improper aiignment
can adversel y affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript and
there are missing pages, these will be noted. Al%, if unauthorized copyright
material had to be removed, a note will indicate the deletion.
Oversize materials (e-g., maps, drawings, charts) are reproduœd by sedioning
the original, beginning at the upper lefi-hand corner and continuing from left to
right in equal sections with small overlaps. Each original is also photographed in
one exposure and is induded in reduced fom at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6" x 9" black and white photographic
prints are available for any photographs or illustrations appeafing in this copy for
an additional charge. Contact UMI directly to order.
Bell & Howell Information and Leaming 300 North Zeeb Road, Ann A b r , MI 481 06-1 346 USA
800-521 -06 O0
DIFFERENTIAL RESPONDING TO THE INTERNAL AND EXTERNAL FACIAL FEATURES:
HOLISTIC AND PART-BASED PROCESSING
Margaret C. McKinnon
A thesis submitted in conformity with the requirements for the degree of Masters of Arts
Graduate Department of Psychology University of Toronto
O Copyright by Margaret C. McKinnon, 1998.
National Library m*! of Canada Bibliothèque nationale du Canada
Acquisitions and Acquisitions et Bibliographie Services services bibliographiques
395 Wellington Street 395. nie Wellington OttawaON KIAON4 Ottawa ON K1A ON4 Canada Canada
The author has granted a non- exclusive licence allowing the National Library of Canada to reproduce, loan, distribute or sell copies of this thesis in microform, paper or electronic formats.
The author retains ownership of the copyright in this thesis. Neither the thesis nor substantial extracts h m it may be printed or otherwise reproduced without the author's permission.
L'auteur a accordé une licence non exclusive permettant a la Bibliothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/film, de reproduction sur papier ou sur format électronique.
L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.
DIFFERENTIAL RESPONDING TO THE INTERNAL AND EXTERNAL FACIAL FEATURES:
HOLISTIC AND PART-BASED PROCESSING
Margaret C. McKinnon Masters of Arts, 1998
Department of Psychology, University of Toronto
Abstract
We investigated earlier daims (e.g., Moscovitch et al., 1997) that recognition of
upright (same and different) and inverted faces relies upon holistic and part-based
decompositional processing strategies associated with the face- and object-
recognition systems, respectively. Viewers made a same-different judgment of two
faces presented either sequentially or simultaneously on screen. Whereas
responding to upright same faces appeared to rely upon holistic or configurational
processing strategies, responding to upright different and al1 inverted faces was
instead mediated by part-based processing strategies. Moreover, attending to the
intemal features alone was sufficient to form a discrimination judgment for upright
different and al1 inverted faces; the same was true for upright same faces when
attention was focused on the intemal features alone at study. By contrast, viewers
relied upon both the interna1 and external features to form a discrimination judgment
for upright same faces when attention was focused on the entire face at study.
Acknowledgments
I would like to express my sincere appreciation to my supervisor, Morris
Moscovitch, for his many insightiul cornments and suggestions during the
preparation of this thesis. 1 am also indebted to Giampaolo Moraglia, my subsidiary
supervisor, for his valuable comments. In addition, Marilyne Ziegler provided
invaluable assistance in the computer programming of these experiments, as did
Elizabeth Olszewska in scheduling participants. I would also like to thank Jason
Gold and Filippo Speranza for their aid in preparing stimuli for an adjunct project
associated with this thesis and Sarah Khan for her assistance in running
participants and analyzing the current experiments.
I would also like to thank my fellow graduate students, Amy Siegenthaler,
Myra Fernandes, Tara Moroz, and Derryn Jewell for providing much appreciated
moral support throughout the year. I am also indebted to Glenn Schellenberg for the
early interest he took in rny career and for his continued encouragement. As always,
I also appreciate the support and encouragement provided by Julia Nelson and
Carlo Sullo. Finally, I thank my family, Matthew McKinnon and Ellen Trinh, and most
especially my parents, Barry and Susan McKinnon, for their love, support, and
guidance, not only in the period surrounding the preparation of this thesis, but
throughout.
Table of Contents
........................................................................................................ List of Tables vi ...................................................................................................... List of Figures vii
General Introduction ..................................................................................... 1
Expe riment 1 ....................................................................................................... M ethod ...................................................................................................... . . .................................................................................... Participants
...................................................................................... Apparatus ............................................................................................ Stimuli
...................................................................................... P roced u re ........................................................................................................ Resuits
........................................................................................ Accu racy .......................................................................................... Latency
Discussion ................................................................................................
Experiment2 ...................................................................................................... 20 Method ..................................................................................................... 21
................................................................................... Participants 21 Apparatus ..................................................................................... 21 Stimuli ........................................................................................... 21 Procedure ................................................................................... 21
Resutts ....................................................................................................... 22 ....................................................................................... Accuracy 22
Latency .......................................................................................... 23 Discussion .......................................................................................... 25
Experiment 3 ...................................................................................................... Method ..................................................................................................... . . ................................................................................... Parttapants
Apparatus ..................................................................................... Stimuli ...........................................................................................
..................................................................................... Procedu re Resuits .......................................................................................................
....................................................................................... Accu racy Latency ..........................................................................................
Discussion ................................................................................................
...................................................................................................... Experiment 4 Method ..................................................................................................... ... Participants ...................................................................................
..................................................................................... Apparat us ........................................................................................... Stimuli
..................................................................................... P roced u re Results .......................................................................................................
....................................................................................... Accu racy .......................................................................................... Latency
Discussion ................................................................................................
.......................................................................................... General Discussion 40
......................................................................................................... References 48
Tables .................................................................................................................. Table 1 ...................................................................................................... Table 2 ...................................................................................................... Table 3 ...................................................................................................... Table 4 ...................................................................................................... Table 5 ...................................................................................................... Table 6 ...................................................................................................... Table 7 ......................................................................................................
...................................................................................................... Table 8
................................................................................................................ Figures 62 Figure1 ..................................................................................................... 62
List of Tables
Table 1 : Mean Accuracy by Mode of Response and Target Format, Experimentl ................................................................................................... 54
Table 2: Mean Reaction Time by Mode of Response and Target Format, Experiment 1 ...................................................................................................... 55
Table 3: Mean Accuracy by Mode of Response and Target Format, Experirnent 2 .................................................................................................. 56
Table 4: Mean Reaction Time by Mode of Response and Target Format, Experirnent 2 ..................................................................................................... 57
Table 5: Mean Accuracy by Mode of Response and Target Format, Experiment 3 ................................................................................................. 5 8
Table 6: Mean Reaction Time by Mode of Response and Target Format, Experiment 3 ................................................................................................... 5 9
Table 7: Mean Accuracy by Mode of Response and Target Format, Expenment 4 ................................................................................................ 60
Table 1: Mean Reaction Time by Mode of Response and Target Format, Experiment 4 ...................................................................................................... 6 1
List of Figures
Figure 1 : Examples of experimental conditions ....................................... 62
DIFFERENTIAL RESPONDING TO THE INTERNAL AND EXTERNAL FACIAL FEATURES:
HOLISTIC AND PART-BASED PROCESSING
General Introduction
Claims of a double dissociation between face and object recognition (e.g.,
Bentin, Allison, Puce, Perez, & McCarthy, 1996; Farah, 1990) have provided an
impetus for the study of facial recognition as distinct from other foms of recognition
(Newcombe, Metha, & de Haan, 1994). In fact, the selective nature of facial
recognition processing has been demonstrated through a variety of experimental
tasks (e.g., functional neuroimaging; Allison et al., 1994; facial inversion; Bartlett &
Searcy, 1993; Searcy and Bartlett, 1996) using both normal (Haxby et al., 1 994)
and brain-damaged (Moscovitch, Winocur, & Behrmann, 1997) human populations,
as well as animal species (Desimone, 1991; Yamane, Kaji, & Kawano, 1988), as
subjects. Moreover, research with young infants (Fagan, 1973) and children
(Peters, 1987) shows that they can discriminate reliably arnong different facial
stimuli. The selective nature of facial processing and the early onset of
discrimination suggests that processing faces is distinct from processing other
complex visual patterns.
Evidence of dissociations between face and object processing is exhibited
in a variety of experimental tasks. For example, recent studies involving functional
neuroimaging and ERP techniques confirrn that faces and objects differ in both the
pattern and location of electrophysiological and blood flow responses that they
evince (e.g., Allison et al., 1994a; 199413). In general, the amplitude of some ERP
waveforms is greater and latency is shorter to faces than to objects, with faces and
objects activating distinct loci in the same general location, usuaily the right
fusifon gyrus. In one fMRl study, McCarthy and his colleagues (McCarthy, Puce,
Gore, & Allison, 1994) reported that presentation of faces resulted in activation of a
srnall region located primarily in the right lateral fusiform gyrus. Similar findings
have been reported in studies involving monkeys as the subject population; these
studies (see for example, Bruce, Desimone, & Gross, 1981 ; Yamane, Kaji, &
Kawano, 1988) confirm that cells within the inferotemporal cortex respond
selectively to faces.
Neuropsychological studies of patient populations also lend support to
claims of a double dissociation between face and object processing, indicating that
these processes may be subserved by distinct neural substrates (Newcombe,
Metha, & de Haan, 1994). Whereas severe deficits in facial recognition
(prosopagnosia) are typically associated with bilateral damage in the region of the
fusiform gyrus, but can be elicited following rig ht-hemisphere damage alone
(Farah, IWO), impaired object recognition is usually linked to bilateral damage in
the same region, but can occur following damage to the left hemisphere alone. In
cases of impaired object or facial recognition, typically the opposing form of
processing is selectively spared (e.g., impaired facial recognition and relatively
intact object processing). The finding that such patients sornetirnes are able to
respond with detailed information when questioned regarding the attributes of
faces or objects that they fail to recognize confirrns that these deficits can occur at
the level of perceptual representation, rather than memory (Banich, 1997).
An additional line of evidence concerning possible dissociations between
face and object processing concems the differential effects of inversion on
recognition of faces and objects. The greater detrimental effect of inversion upon
face- than object-recognition is well-docurnented (e.g., Bartlett & Searcy, 1 993;
Diamond & Carey, 1986; Rhodes, Brake, & Atkinson, 1993; Searcy & Bartlett, 1996;
Yin, 1969). Specifically, inversion of faces relative to the viewer results in impaired
encoding; this effect is greater for faces than for most, but not al1 (Diamond & Carey,
1986; Yin, l969), other objects. Moreover, although dissenting opinions have
been expressed (e.g., Valentine, 1988, 1 W l ) , the available evidence (Bartlett &
Searcy, 1993, Rhodes et al., 1993; Searcy & Bartlett, 1996) suggests that inversion
has a greater effect upon the encoding and subsequent discrimination of the
spatial-relational (i.e., configurational) information contained in facial stimuli than it
does on discrimination of isolated facial components (e.g., eyes, ears, and mouth).
Indeed, several researchers maintain (see for example, Carey, 1978; 1981 ; Farah,
Tanaka, & Drain, 1995; Rhodes et al., 1993; Tanaka & Farah, 1993; Young,
Hellawell, & Hay, 1987) that whereas recognition of upright faces may rely upon
holistic or "configurational" (see Moscovitch et al., 1997; Yin, 1969) representation
schemes, recognition of inverted faces, much like objects, may rely upon part-
based decompositional strategies (see Marr & Nishihar, 1978 for an original
interpretation, as well as Biederman & Gerhardstein, 1993). Hence, recognition of
faces and objects may differ according to the degree of part-based decomposition
required for processing of the two types of stimuli, with recognition of upright faces
requiring little such decomposition or attention to component features of the face,
such as the eyes, nose, or mouth. Because faces are typically presented upright,
recognition performance for inverted faces may suffer as a result of the unfamiliar
part-based representations that accompany inversion (Haman & Moscovitch, in
preparation; Farah et al., 1995).
Several recent studies by Farah and her colleagues provide partial support
for these claims. Specifically, Tanaka & Farah (1 993) found that participants
performed more poorly when required to recognize isolated parts of upright faces
than when required to recognize the same components of inverted or scrarnbled
faces and objects. A later study (Farah, Tanaka, & Drain, 1995) revealed that
inversion effects are also apparent for non-face stimuli if participants are
encouraged to encode these stimuli as whole units or perceptual "gestalts", rather
than decompose them into parts. By contrast, treating faces like objects appears to
lessen the effects of inversion. Specifically, participants in the Farah et al. (1995)
study were required to study line drawings of unfamiliar faces in either whole form
or decomposed into parts. At test, participants were required to identify whole
faces presented either upright or inverted; inversion of the test faces resulted in
greater deficits when faces were studied as wholes rather than as parts. Hence,
initial representation of complex stimuli in a holistic form appears to result in
impaired recognition performance at inversion. The detrimental effects of inversion
are not as apparent, however, when stimuli are initially represented in a part-based
manner. A recent series of experiments by Harman and Moscovitch (in
preparation) replicated these findings using fractured (part-based) and intact
(upright and inverted) photographs of whole faces, while also suggesting that
format incongruency between study and test stimuli may have contributed to the
inversion effect. Taken together, the results of the studies by Farah and her
colleagues, as well as those of other researchers (Carey, 1978; 1981 ; Harman &
Moscovitch, in preparation; Moscovitch et al., 1997; Rhodes et al., 1993; Young,
Hellawell, & Hay, 1987) constitute strong evidence for daims of a dissociation
between face and object processing by suggesting that these processes may rely
upon two distinct mechanisms, one that is specialized for holistic or configurational
processing, and another that relies upon part-based representations.
Although dissenting opinions have been expressed (see for example,
Esgate, Burton, & Burton, 1996; Kreuger, 1978), the common observation voiced in
much of the literature is that whereas responding for same items may rely upon
rapid, global modes of processing, different responses may instead reiy on more
analytic or feature-based cornparisons (Bagnara, Boles, Simion, & Umiltà,, 1982;
Hock, 1973; Tzylor, 1 976a, 1976b). This pattern of responding may extend to the
processing of facial stimuli. For example, in one early study, Smith and Neilson
(1970) required participants to make a same-different judgment of schematic line
drawings of faces. At delays of 1 to 4 seconds, the number of features differing
between the two faces predicted the latency of different responses, with less tirne
being required for judgments of faces comprised of a greater number of different
features. Such effects were not apparent for same judgments, however; these
judgments were not affected by the total number of features present in the faces,
suggesting that participants did not engage in a serial comparison of the individual
features for these judgments, as was likely upon different responses. At delays of
10 seconds, however, both same and different judgrnents yielded a pattern of
results consistent with a reliance upon feature-by-feature comparison processes.
Although these findings (Smith 8 Neilson, 1970), as well as those of other
researchers (Bradshaw & Wallace, 1971), provide some evidence that processing
strategies for faces may Vary as a function of same versus different modes of
responding, with more holistic or configurational processing strategies being
associated with same judgments of facial stimuli, and more analytic or feature-
based comparisons being required for different judgments, particularly at short
retention intervals not requiring long-term or consolidated memory representations
(e.g., Bradshaw & Wallace, 1971 ; Smith & Neilson, 1970), such conclusions are
equivocal given viewers' reliance upon both holistic and feature-based strategies
for same and different judgments other studies of a similar nature (e.g., Sergent
1 984; Matthews, 1978).
The present experirnents will examine these daims in light of one aspect of
facial processing that has received much attention (e.g., Ellis, Shepard, & Davies,
1979; Nachson, Moscovitch, & Umiltà, 1995) in the recent literature: the
relationship between the internal and external facial features. Evidence of a
differential perceptual status for the intemal and externat features of facial stimuli is
seen in early infancy. Infants younger than two months of age typically limit their
visual scan to extemal facial features (Mauer & Salapatek, 1976). By contrast,
infants older than two months shift their gaze to internal facial features;
explanations for this effect have centered upon the optimal spatial frequencies
(Valenza, Simion, Cassisa, & Umiltà, 1996) of the stimuli, as well as their
differential complexity (e.g., Mauer, 1983). There is preliminary evidence
(Campbell & Tuck, 1995) to suggest that the recognition of farniliar faces follows a
similar pattern of development. Specifically, although younger children appear to
rely more heavily upon the external facial features for recognition, by age 10
recognition of familiar faces is achieved more efficiently by presentation of internal,
rather than external, facial features.
Other researchers (de Haan & Hay, 1986; de Haan, Young, & Newcornbe,
1987; Ellis et al., 1979; Hines, Jordan-Brown, & Juzwin, 1987; Moscovitch et al.,
1997; Nachson et al., 1995; Young, Hay, McWeeny, Flude, & Ellis, 1985) have
investigated the differential contribution of the internal and external features to the
recognition of familiar and unfamiliar faces in regards to adult populations. For
example, Ellis and his colleagues (Ellis, Shepard, & Davies, 1979) assessed
participants' reliance upon intemal and external features for identification.
Specifically, participants were requi red to identify familiar and unfamiliar faces
upon brief presentations of either the internal or external features. The results of
their study indicated that whereas participants appear to rely heavily upon the
information available in the internal facial features for identification of familiar faces,
there was no reliable difference in identification rates for unfamiliar faces when
participants were presented with internal or external features.
Later studies (de Haan & Hay, 1986; Hines, Jordan-Brown, & Juzwin, 1987;
Nachson et al., 1995; Young et al., 1985; see Moscovitch et al., 1997 for a different
experimental method) using matching tasks have revealed conflicting results
regarding the contribution of intemal and extemal features to the recognition of
familiar and unfamiliar faces. For example, whereas two studies (de Haan & Hay,
1986; Nachson et al., 1995) that used different experimental methods provided
evidence that external features are more efficacious than internal features in
matching tasks for unfamiliar faces, other studies [Ellis et al., 1979 (identification
task); Hines et al., 1987; Young et al., 19851 have failed to replicate this pattern of
findings. By contrast, the results of Young et al. (1985), as well as other
researchers (Campbell & Tuck, 1995; de Haan & Hay, 1986; Ellis et al., 1979)
converge on the finding that recognition of familiar faces is best achieved through
presentation of the intemal features.
Neuropsychological studies have also dernonstrated a pattern of diff erential
responding to internal and external facial stimuli. Ptischel and Zaidel (1994)
reported that removal of external facial features affected performance in a same-
different task when stimuli were presented to the right (left hemisphere) but not left
(right hernisphere) visual field. Hence, the left hemisphere may be less dependent
upon feature-dependent mechanisms, such as those found in the intemal features,
than is the right hernisphere. In a seemingly contradictory pattern of findings,
however, Young (1 984) reported a right-hemisphere superiority for recognition of
familiar faces from both internal and external features. This pattern of responding
was replicated in a later study by deHaan and Hay (1986) involving both normal
controls and patients with diffuse damage to either the left (LH) or right (RH)
hemisphere as the subject population. Normal controls exhibited a right-
hemisphere (left visual field) advantage for a sarne-different task requiring
participants to make judgments between a whole face and a face comprised of the
intemal or external features only. ln addition, patients suffering RH damage were
found to exhibit greater functional impairment for this task than were patients with
LH damage. A later neurophysiological study with monkeys as the subject
population (Horel, 1 993) indicated that bilateral cooling of the inferotemporal cortex
resulted in a severe irnpainent of monkeys' ability to perform a task that required
them to discriminate among different within-species faces that were identical
except for their internal features. Hence, although there is preliminary evidence of
a possible neurological (as well as functional) dissociation between the internal
and external features, the current status of this claim is equivocal at best.
Although there is preliminary evidence of a differential perceptual status for
the processing of internal and external features of facial stimuli, further
investigation is required to clarify the nature of the dissociability. Indeed, these
features may subsume distinct roles in facial recognition. A recent study by
Moscovitch and his colleagues (Moscovitch et al., 1997) that demonstrated
selective sparing of recognition w hen extemal featu res were inverted lends partial
support to this claim. Moscovitch and his colleagues recently compared facial
recognition in a patient (CK) with impaired object recognition, but intact facial
recognition processing, (Le., associative visual object agnosia and dyslexia; see
Behrmann, Winocur, & Moscovitch, 1992; Behnnann, Moscovitch, & Winocur, 1994
for a discussion in regards to this patient), to that of normal controls. Indeed, the
results of this research demonstrated that recognition of fractured and inverted
faces may rely upon the object-recognition system that is thought to be impaired in
this patient. In one experiment, the effects of inversion of the intemal versus
extemal features on recognition of a familiar face were compared. The results of
Moscovitch et al.'s experiment indicated that although inversion of the interna/
facial features resulted in irnpaired performance for both normal controls and CK,
the observed impairment in recognition was much greater for CK (drop of 60
percent versus drop of 20 percent in normal controls). By contrast, inversion of the
extemal features had no effect on recognition for either CK o r the controls. A
su bsequent experiment also resulted in impaired recognition performance when
the spatial relations among the internal features of a fractured face were disrupted;
again the detrimental effects of this manipulation were greater for CK than for
controls, suggesting that control subjects were able to rely on other mechanisms,
such as those used to identify objects, to aid in recognition.
Although it is possible that under conditions of inversion recognition is
subsumed by the individual features alone, Moscovitch and his colleagues
(Moscovitch et al., 1997) present recent evidence that renders this conclusion
unlikely. Specifically , the patient CK who demonstrated intact facial recognition,
but impaired object recognition, was able to recognize individual facial features as
well as controls, an indication that the facial-recognition system (which apparently
suffers breakdown under conditions of inversion) is attuned to these features. In
cases where CK was able to identify successfully inverted or fractured faces, he
sometimes appeared to do so through identification of separate features that were
especially salient (e.g., Prince Charles' large ears). Moreover, when he had
difficulty recognizing inverted cartoon faces, as well as overlapping objects and
Mooney figures (black and white pictures of objects and faces that are constructed
of patches of intense light and shadow), it appeared to be because he had difficulty
in appreciating the relation of the parts to the whole, an impairment that may have
hindered reintegration. Similarly. it is unlikely that facial recognition under
conditions of inversion is reliant upon mental rotation of inverted stimuli to an
upright position (Bartlett & Searcy, 1 993). Mental rotation of corn plex stimuli,
including faces, is impossible or highly error-prone (Rock, 1973). Hence, partial
reintegration or "normalization" of the disrupted facial features, as postulated by
Harman and Moscovitch (in preparation), as well as other researchers (Farah et al.,
1995), appears a likely possibility for recognition following inversion or other such
disruptions of facial processing.
Because inversion of the internal features affected recognition (as did
disrupting the spatial relations among the internal features of the fractured faces),
but inversion of the external features did not (Moscovitch et al., 1997), it appears
likely that the intemal features may carry the "burden of information" (p. 575) in
normal facial recognition. The relative importance of the different features under
conditions of impaired facial recognition processing (e.g., inversion), however, is
less clear. In the experiment conducted by Moscovitch and his colleagues, the
internal and external features were manipulated separately, yet in the context of a
whole face. For example, when the intemal features were inverted, the extemal
features remained upright and vice versa. Further research is necessary to clarify
the differential contribution of the internal and external features under conditions
that manipulate presentation of these features (e.g., full congruency of inversion).
One possibility suggested by Moscovitch and his colleagues is that the external
features (or facial contours) may be used as a frame or point of reference for
realignrnent of the internal features when they are impaired or fractured, acting as a
sort of perceptual "anchor" for reintegration or normalization. Indeed, several
researchers (Farah et al., 1995) have suggested that under conditions of inversion,
the locus of processing switches from that of the facial-recognition system to the
object-recognition system. Because it is apparent that the object-recognition
system has access to the extemal features (Moscovitch et al., 1997), it is possible
that it uses these features to guide reintegration. In fact, a similar proposal has
been made by Hurnphreys and his colleagues (Humphreys et al., 1994) in regards
to the perceptual reintegration of objects.
In the present series of experiments, we proposed to investigate differential
processing of the internal and external facial features through a series of
manipulations involving sarne-different judgments of unfamiliar faces. We
investigated these claims using an inversion paradigm by initially presenting
participants with upright faces with the external features either present (whole face
condition) or absent (internal features only) and later testing their ability to rnake
same-different judgments between these faces and a cornparison face that was
presented either upright or inverted. Whereas recognition of upright faces rnay be
subsumed by the facial-recognition system that relies upon holistic or
configurational processing strategies, the available evidence suggests that
recognition of inverted faces rnay be mediated by the object-recognition system
that relies upon analytic or part-based decompositional strategies for discrimination
(Moscovitch et . al., 1997). Although holistic or configurational processing
strategies required for the processing of upright faces rnay dictate attention to the
entire face, a reliance upon more part-based decompositional strategies requiring
attention to the individual facial features (e.g., eyes, nose, and mouth) rnay facilitate
responding to the internai features alone. We rnanipulated viewers' initial attention
to such features by presenting either an upright whole face or a part face at study.
Nonetheless, such strategies rnay be additionally mediated by responding to
upright same versus upright different cornparison items. Specifically, the results of
several earlier investigations (Bradshaw & Wallace, 1971 ; Matthews, 1978; Smith
& Neilson, 1970; Sergent, 1984) provide some preliminary evidence that whereas
responding to upright same items rnay rely upon holistic or configurational
processing strategies, responding to upright different items rnay rely upon analytic
or part-based decornpositional processing schemes. Although the available
evidence strongly suggests that the facial-recognition system relies predominately
upon holistic representation schemes to mediate recognition of upright faces
(Farah et al., 1995; Moscovitch et al., 1997), it is possible that performance for
same and different items requires this system to accommodate both holistic and
part-based representational schemes, respectively. By contrast, we did not expect
viewers' performance for inverted faces to be mediated by same versus different
modes of responding, instead relying chiefly upon part-based decomposition
schemes dictated by the object-recognition system (Biederman & Gerhardstein,
1993; Marr & Nishihar, 1978; Moscovitch et al., 1997).
Because the intemal features are thought to carry the burden of information
in facial recognition (Moscovitch et al., 1997), attention to these features may be
sufficient to f o m a discrimination judgment. Although recent evidence
accumulated by Moscovitch and his colleagues with the patient CU suggests that
this patient's difficulty in recognizing inverted (as well as fractured) faces may have
arisen from his inability to appreciate the relation of parts of these stimuli to the
whole (i.e., reintegration of the face to a perceptual gestalt), it is possible that
successful recognition performance following inversion is possible by attending to
the intemal features alone. An altemate possibility we considered was that the
preferential access accorded the external features by the object-recognition system
following inversion may allow these features to serve as an anchorhg point for
reintegration or normalization of disrupted facial stimuli. Because recognition of the
external features is not thought to be impaired following inversion (Moscovitch et
al., 1997), reintegration or normalization of facial stimuli may be required for the
internal features only. Hence, an additional aim of the current study was to
determine whether the internal and external features played a differential role in
discrimination performance upon upright and inverted presentations, perhaps as a
result of different processing strategies required for the discrimination of upright
(same and different) and inverted faces (Farah et al., 1995).
Experiments I and 2: lnitial Presentation of Whole Faces
Experiment 1 : Simultaneous Discrimination Task lnvolving Initial Presentation of
Whole Faces
Introduction. In the first experiment, participants were required to make a same-
different judgment of a test and comparison face presented simultaneously on
screen. Study faces were presented upright with both the internal and external
features present (whole face condition). By contrast, comparison faces were
presented either upright or inverted. Half of the comparison faces were presented
with both the internal and extemal features present (whole face); the remaining
faces were comprised of the intemal features only (part face). Because inversion of
faces relative to the viewer results in impaired perceptual encoding (see for
example, Bartlett & Searcy, 1993; Yin, 1969), this paradigm allowed us to compare
participants' ability to reliably discriminate different facial stimuli under conditions of
intact (upright) and impaired (inverted) facial processing.
Whereas recognition performance for inverted faces is thought to rely upon
part-based processing strategies mediated by the object-recognition system that
rnay require attention to individual parts of faces (Moscovitch et al., 1997),
recognition performance for upright faces rnay instead rely upon holistic processing
strategies requiring attention to the entire face. Moreover, the available evidence
(Bradshaw & Wallace, 1971 ; Smith & Neilson, 1970) suggests that performance for
same versus different upright faces rnay be additionally rnediated by holistic and
part-based decompositional processing strategies, respectively. Because initial
study presentations involved an upright whole face, we expected that participants
would rely upon holistic or configurational processing strategies at study. Indeed,
we expected these processing strategies to result in superior performance for the
presentation of whole rather than part same comparison faces that would also rely
upon holistic processing strategies for discrimination. By contrast, presentations of
whole comparison faces would no longer be favored upon upright different and al1
inverted comparison presentations, when discrimination performance rnay be
based upon individual consideration of the internal and extemai facial features.
One possibility we considered was that part-based processing strategies
associated with the recognition of inverted faces rnay lead to a reliance upon the
intemal features alone (part face condition) following inversion. Indeed, because
the intemal features are thought to carry the burden of information in facial
recognition, attention to these features alone rnay be sufficient to form a
discrimination judgment. If, however, the extemal features serve as an anchoring
point for reintegration following inversion (Moscovitch et al., 1997), we might expect
superior performance for inverted presentations of whole (intemal and external
features), rather than part, comparison faces following inversion. Because this
experiment involved sirnultaneous presentations of only two faces, responding
relied chiefly upon participantst perceptual representations of the stimuli, and was
unlikely to engage the rnemory systems directly.
Method
Participants. The participants were 24 members of the University of Toronto
at Mississauga comrnunity; five were male and 19 female. Al1 had either normal or
corrected-to-normal vision. Participants received nominal payment or academic
credit for their participation in the experirnent, which took approximately 20
minutes.
Apparatus. The stimuli were 16 unfamiliar faces, half male and half female,
taken from yearbook photographs and scanned into a computer file by a software
scanning prograrn (Picture Publisher 4.0, Micrografx) installed on a Computech
486-33 PC computer. Stimulus presentation and response recording were
controlled by a customized software program created by a commercially-produced
software program (MEL 2.0). Stimuli were presented via the computer screen on a
15-inch View Sonic 15GS monitor; participants used a response box connected to
a Pentium 166 PC computer to record their responses.
Stimuli The stimuli were unfamiliar faces (standards) scanned at a constant
level of contrast. Ail provided a frontal view of the face; sideways views were not
included (experimenter's judgment). The final size of the scanned images was 180
pixels wide by 250 pixels in height. Along with the study faces (upright position;
whole face), four different sets of comparison faces served as stimuli: i) upright
faces (whole face); ii) upright faces (intemal features only); iii) inverted faces
(whole face) and; iv) inverted faces (internal features only) (see Figure 1). The
intemal features of the face, consisting of the eyes, nose, and mouth, were cut out
with comparable pentagon-shaped tracing lines. Sixteen (eight male, eight female)
different standard faces were used to forrn the test and comparison faces. Each of
the sixteen standard faces seived as a study face and was manipulated to satisfy
the requirements of the four comparison conditions for a total of 64 comparison
faces. On each trial, the external features (when present) were identical across
study and comparison faces. Four different sets of external features were used.
Hence, four male faces shared the same external features; the remaining four male
standards shared a different set of external features, The same was true of the
female standards. Faces were matched such that when different cornparisons
were made, the internal features of the study and comparison faces were as similar
as possible (e.g., similarly shaped eyebrows or mouths). The stimuli were
presented as Greyscale (black-and-white) images on a white background.
---4 nsert Figure 1 about here-----
Each comparison face was presented twice (once as a same face and once
as a different face) alongside a study face in a different random order for each
participant for a total of 128 test trials. Sixteen practice trials (eight sarne, eight
different) were constructed from a different set of male faces. The practice trials
were identical to the test trials, with the exception that on different trials, the internai
features of the comparison faces differed more widely (e.g., incongruent shape of
mouth) from the study faces than was true on actual experimental trials.
Procedure. Participants were tested individually and received instructions
both verbally and on the cornputer screen. Their task was to make a same-different
judgment of a study and comparison face based on the internal features only.
Same, in this experiment, referred to full congruency between the internai features
of study and comparison faces; participants were instructed that differences
(including omission) between the external features of the study and comparison
faces were not to be considered in their judgment. At conclusion of the testing
session, few participants reported noticing that the external features were constant
(i.e., identical) across study and comparison presentations. Participants were
instructed to respond as quickly and as accurately as possible; both accuracy and
response-time data were recorded. Al1 trials were self-paced.
To eliminate possible hemispheric biases in responding, stimuli were
centered on screen, with study and cornparison faces vertically aligned. Study
faces consisting of both the interna1 and external facial features (whole face
condition) were presented at the top of the screen; comparison faces were
presented at the bottom of the screen. All trials (including the practice trials)
involved simultaneous presentations of the study and comparison faces. Stimuli
remained on screen until the initiation of a response by the participant. Following
responding, a fixation cross appeared on screen signaling participants to initiate a
new trial. Hand of responding was counterbalanced across participants.
Participants initially completed sixteen practice trials (eight same, eight
different), during which they were allowed to ask questions about the experimental
procedure. The actual test session consisted of 128 trials; sixty-four trials were
same trials, the remaining 64 were different trials. The test trials were presented in
a different random order for each participant.
Results.
Accuracy. Inversion of test faces resulted in decrements in viewer
performance. When the response data were analyzed by grouping responses for
same versus different presentations, however, it was revealed that holistic
processing strategies contributed to marginally (though not statistically significant)
superior performance upon upright same presentations of whole comparison faces
only (see Table 1 ). Hence, although differences in performance for upright and
inverted faces may have stemmed from either differences in format between test
and study faces (Harman & Moscovitch, in preparation) or in encoding strategies
used by viewers (Farah et al., 1995), accuracy scores exhibited little evidence of
the effects of holistic versus part-based processing strategies on discrimination of
part versus whole comparison faces.
----- lnsert Table 1 about here ----- These results were confirmed using a factorial ANOVA design. Means were
entered into a 2 (Mode of Responding: Same and Different) X 2 (Target Orientation:
Upright and Inverted) X 2 (Target Format: Part and Whole) factorial ANOVA with
Mode of Responding, Target Orientation, and Target Fona t treated as within-
subjects factors. The results of testing revealed a main effect of Target Orientation
(E(1, 23) = 1 17.64, g c .O01 ; upright faces were processed more accurately than
were inverted faces. Main effects of Mode of Responding also approached
significance, (E(1, 23) = 3.19, g c .09; viewers exhibited superior performance for
same than for different responses. Main effects for Target Format (E (1, 23) = .32, g
> .05) were non significant; there were no interactions between Mode of
Presentation and Target Orientation (E (1, 23) = 0.00, g > .05), Mode of
Presentation and Target Form (E (1 , 23) = 0.05, g > .05), or between Target
Orientation and Target Format (E (1, 23) = 1.76, g > .05). The three-way interaction
between Mode of Responding, Target Orientation, and Target Form also failed to
attain significance, (E (1, 23) = 1.29, g 2 .05).
Post-hoc comparison contrasts between faces presented upright and faces
presented inverted using the Ryan-Einot-Gabriel-Welsch multiple range q-test
(REGW-Q) confimed that discrimination performance for faces presented upright
was better than that for faces presented inverted (g c -05).
Latencv.
The results for the latency data were similar to those reported for the
accuracy data in that response time was faster for faces presented upright than it
was for faces presented inverted. When these data were analyzed by grouping
responses by mode of presentation (i.e., same versus different), however, it was
revealed that viewers exhibited superior performance for upright same
presentations of whole faces. By contrast, at inversion viewers exhibited little
differences in responding for part versus whole presentations of comparison faces,
regardless of the mode of responding required (see Table 2). No differences were
apparent in responding to whole versus part upright different comparison faces.
----- lnsert Table 2 about here -----
These results were confirmed using a 2 (Mode of Responding: Same and
Different) X 2 (Target Orientation: Upright and Inverted) X 2 (Target Format: Part
and Whole) factorial ANOVA with Mode of Responding, Target Orientation, and
Target Format treated as within-subjects factors. Medians were calculated for each
subject for each condition; only reaction times for accurate responses were
included in the analysis. In addition, we elirninated from the analysis al1 reaction
times that fell more than two standard deviations frorn the mean and calculated
new medians using the remaining data (see Ratcliff, 1993 for a discussion of the
treatment of reaction time outliers). This procedure resulted in elimination of less
than 4.3 percent of the total observations. The results of testing revealed a main
effect of Target Orientation (E(1, 23) = 21 -21, Q < .O01 ; upright faces were
processed more efficiently than were inverted faces. This result was confirmed by
post-hoc analysis using the REGW-Q (g < .05). There was also an interaction
between Mode of Responding and Test Orientation (E(1, 23) = 4.50, g c .OS) such
that differences in responding between upright and inverted faces were less
apparent for different (E(1, 23) = 38, g > .05) than for same (E(1, 23) = 4.03, g c .06)
responses. A three-way interaction between Mode of Responding, Target
Orientation, and Target Format (E (1, 23) = 577, p c .05) confimed that whereas
responding for upright same items varied as a function of part versus whole
presentations (E (1, 23) = 10.03, g < .01), responding did not Vary for either upright
different responses (E (1, 23) = 0.02, g > .05) or inverted same (E (1,23) = 1.41, g >
.05) and different (E (1, 23) = 0.37, g > .05) responses. Hence, viewers exhibited
superior performance for whole, rather than part, presentations of comparison
faces upon upright same presentations only. Main effects for both Mode of
Presentation (E (1 , 23) = .24, > .05) and Target Format (E (1, 23) = .29, g > -05)
were non significant; there were no interactions between Mode of Presentation and
Target Form (E (1, 23) = 2.58, g z .05) or between Target Orientation and Target
Format (E (1, 23) = 0.06, g > .05).
Discussion.
Faces presented upright were recognized more efficiently than were
inverted faces. In addition, whereas the results of the accuracy analysis provided
little evidence that responding varied as a function of part versus whole
presentations, the resuits of the latency analysis were clear in suggesting that
viewers exhibited superior performance for whole, rather than part, presentations of
upright sarne cornparison items. We speculate that such performance may arise
from a reliance upon holistic encoding strategies required for recognition of upright
same faces; these strategies may require attention to the entire face upon such
presentations.
By contrast, holistic encoding strategies resulting from study presentations of
an upright whole face may be incongruent with more part-based encoding
strategies required for the processing of upright different faces and al1 inverted
faces. These strategies may result in representation of whole comparison faces as
two separate parts, comprised of the internai and external facial features;
discrimination judgments following such representations may be based upon
individual consideration of the intemal and extemal features, mitigating any
advantages previously observed for whole faces. Indeed, viewers' tendency
toward part-based processing may have been counteracted by initial holistic
representations of the entire face.
We found no evidence that the external features serve as an anchoring point
for the reintegration of facial stimuli following inversion; performance was
equivalent for part and whole presentations of inverted cornparison faces.
Although this result favored our view that the intemal features may provide
sufficient information to form a discrimination judgment following inversion, the
equivalent performance observed for part and whole faces upon inversion
provided little evidence to support this claim. A second experiment, involving rapid,
sequential presentations of response items, was designed to heighten viewers'
initial reliance upon holistic processing strategies for upright study faces.
Experiment 2: Sequential Discrimination Task lnvolving Initial Presentation of
Whole Faces
Introduction.
In the first experiment, participants were required to make a same-different
judgment of a study and comparison face presented simultaneously on screen.
Because stimuli remained on screen until the initiation of a response, participants
may have relied on feature-by-feature comparisons between component parts of
study and comparison faces, rather than holistic or configurational encoding of the
images, to facilitate responding. We reasoned that conditions involving a higher
perceptual load (Le., more rapid encoding) would be more likely to result in
configurational or global processing (Bagnara, Boles, Simion, & Umiltà, 1982) of
upright faces by precluding, in part, the possibility that participants would be able to
rely on feature-by-feature comparisons between test and studied items (as when
stimuli remained on screen until responding) to facilitate responding. Hence, faces
were presented sequentially for a brief duration. In addition, a visual mask,
comprised of jumbled faces not seen in the experiment, was introduced between
presentations of the study and comparison faces; its purpose was to prevent
participants from forming an enduring perceptual representation of the initial study
face.
Although neither experiment 1 nor 2 allowed for an exploration of the effects
of long-terrn memory representation on subsequent discrimination, the rapid nature
of the stimulus presentations in the current experiment required that participants
form a judgment based upon their memory representations of the stimuli, perhaps
in working memory .
Method.
Partici~ants. The participants were 24 members of the University of Toronto
at Mississauga community; six were male and 18 female. All had either normal or
corrected-to-normal vision. Participants received nominal payment or academic
credit for their participation in the experiment, which took approximately 15
minutes.
Amaratus. The apparatus was identical to that in Experiment 1.
Stimuli. The study and comparison faces were identical to those in
Experirnent 1. In addition, a visual mask, comprised of the individual features of
several jurnbled faces, was introduced. The mask, like the study and comparison
faces, was a Greyscale image 180 pixels wide and 230 pixels in height presented
on a white background; it was created by mixing together the component parts of
several different faces not seen in the experiment.
Procedure. The procedure was identical to that in Experiment 1, with the
exception that in order to preclude the possibility of feature-by-feature cornparisons
between study and comparison faces, stimuli were presented sequentially for a
brief duration, with the introduction of a visual mask between study and cornparison
images. On each experimental trial, a study face consisting of both the intemal and
external features (whole face) was presented for 1 second (sec), followed by the
presentation of a visual mask for a 500 rnillisecond (rnsec) duration. lmrnediately
following presentation of the mask, a cornparison face appeared on screen for 1
sec. Stimuli, including the visual mask, were centered on the computer screen. All
trials were self-paced; following presentation of the cornparison face, a fixation
cross appeared on screen at which time participants were instructed to record their
responses and then initiate a new trial. As in Experiment 1, both accuracy and
response-time data were recorded.
Participants initially completed sixteen practice trials (eight same, eight
different) during which they were allowed to ask questions about the experimental
procedure. The actual test session consisted of 128 trials; sixty-four were same
trials, 64 were different trials. Test trials were presented in a different random order
for each participant.
Results.
Accuracv. Performance in the second experiment was somewhat worse
than in the first experiment. Nonetheless, performance for both upright and
inverted faces appeared to be rnediated by holistic and part-based
representational strategies associated with the face- and object-recognition
systems, respectively (Farah et al., 1995; Moscovitch et al., 1997), as well as
differences in processing strategies for same and different facial stimuli (upright
faces only) (see Table 3).
-----Insert Table 3 about here-----
These results were confirmed using a factorial ANOVA design. Means were
entered into a 2 (Mode of Responding: Same and Different) X 2 (Target Orientation:
Upright and Inverted) X 2 (Target Format: Part and Whole) factorial ANOVA with
Mode of Responding, Target Orientation, and Target Format treated as within-
subjects factors. The results of testing revealed a main effect of Target Orientation
(E(1, 23) = 117.64, g c .O01 and Mode of Responding (E(1,23) = 16.98, g c .O01 ;
upright faces were processed more accurately than were inverted faces and same
faces were processed more accurately than were different faces. A two-way
interaction between Mode of Responding and Target Format was also observed, (E
(1, 23) = 15.38, g < .OOl); planned contrasts confirmed that although performance
for upright faces differed as a function of same versus different modes of
responding (E (1, 23) = 49.09, g c .001), performance for upright faces did not (E
(1, 23) = 2.13, p < .05). The two-way interaction between Target Orientation and
Target Format also attained significance (E (1, 23) = 7.42, g < .02); a simple effects
analysis revealed that whereas viewers exhibited superior performance overall for
upright presentations of whole faces (E (1, 23) = 196.1, g < .001), superior
performance was observed for inverted presentations of part faces (E (1,23) =
103.0, p < Bol). Main effects of Target Format (E (1, 23) = 0.00, g > .05) were non
significant; there were no interactions between Mode of Presentation and Target
Orientation (E (1, 23) = 0.93, p > .OS), or between Mode of Responding, Target
Orientation, and Target Fomi (E (1, 23) = 2.95, > .05).
Post-hoc cornparison contrasts using the REGW-O between faces presented
upright and faces presented inverted and between same and different
presentations confirmed that discrimination performance for faces presented
upright was better than that for faces presented inverted and for same than for
different responses (es < .05) . Latencv. When the latency data were analyzed by grouping responses for
same versus different presentations, performance for inverted faces was somewhat
better for whole than for part faces, regardless of the mode of responding (i.e.,
same versus different) required (see Table 4). By contrast, performance for upright
faces was again rnediated by same versus different modes of responding.
Whereas performance for upright, same faces was best when comparison faces
were whole faces, performance for upright, different faces was best when
comparison faces were comprised of only part of the face (i.e., the intemal
featu res).
-----Insert Table 4 about here-----
These results were confirmed using a 2 (Mode of Responding: Same and
Different) X 2 (Target Orientation: Upright and Inverted) X 2 (Target Format: Part
and Whole) factorial ANOVA with Mode of Responding, Target Orientation, and
Target Format treated as within-subjects factors. As in Experiment 1, medians were
calculated for each subject for each condition; only reaction times for accurate
responses were included in the analysis. In addition, we eliminated from the
analysis al1 reaction times that fell more than two standard deviations from the
mean and calculated new medians using the remaining data. This procedure
resulted in elirnination of less than 3.7 percent of the total observations. Testing
revealed a main effect of Target Orientation (E(1, 23) = 30.65, g < .001); upright
faces were processed more efficiently than were inverted faces. This result was
confirrned by post-hoc analysis using the REGW-Q (e c .05). There was also a
main effect of mode of responding (E(1, 23) = 7.87, Q < .O2 ), with faster responses
being made for same than for different presentations (REGW-Q, g < .05). A three-
way interaction between Mode of Responding, Target Orientation, and Target
Format (E (1, 23) = 5.77, g < .05) confirmed that viewers exhibited the greatest
differences in performance between whole and part presentations of comparison
faces following upright same presentations (E (1, 23) = 10.03, g c .01). Main effects
of Target Format (E (1, 23) = 2.1 1, g > .05) were non significant; there were no
interactions between Mode of Presentation and Target Orientation (E (1, 23) = 1.47,
Q > .05), Mode of Presentation and Target Form (E (1, 23) = 2.75, p > .05), or
between Target Orientation and Target Format (E (1, 23) = .48, g z .05).
Discussion. The results of Experiment 2 both confinned and extended our earlier
observations in Experiment 1. The main finding of both experiments was that
differential processing strategies associated with the recognition of upright and
inverted faces (Farah et al., 1995; Moscovitch et al., 1997), as well as responding to
same versus different upright faces (Bradshaw & Wallace, 1971 ; Matthews, 1978;
Smith & Neilson, 1 970; Sergent, 1 984), appear to rnediate viewers' responses to
facial stimuli.
Performance in Experiment 2 was best for whole, rather than part,
presentations of upright same faces. By contrast, whereas viewers exhibited more
accurate performance for whole presentations of upright different faces, response
latency was actually faster for part presentations of these faces, suggesting a
speed-accuracy trade-off; the more accurate responses observed for whole faces
upon different responses may reflect the longer amount of processing time
accorded these stimuli (i.e., responses were approximately 83 msec slower to
whole than to part comparison faces).
The observed disparities in responding for upright faces in Experiments 1
and 2 may stem from differences in the processing of same and different faces. In
both experiments, viewers exhibited superior performance for same presentations
of whole rather than part comparison faces, likely reflecting a reliance upon holistic
encoding strategies requiring attention to the entire face. By contrast, we found
little evidence of differences in responding to part versus whole faces upon upright
different presentations. We speculate that an initial reliance upon holistic encoding
strategies at study of whole faces may have counteracted more part-based
representations of upright different faces, that may have relied upon separate
consideration of the intemal and extemal features. Hence, we observed few
differences in performance for part versus whole comparison faces upon such
presentations.
The results of the accuracy and latency data for inverted faces in Experiment
2 were somewhat contradictory in suggesting that whereas viewers exhibited more
accurate performance for part rather than whole presentations of companson faces,
efficiency was sornewhat better for whole than for part presentations. Most notably,
performance was approximately 100 msec faster for whole presentations of
inverted different faces, while responding in the same condition was more accurate
for part faces. Nonetheless, the low viewer accuracy (approximately 55 percent
correct, a response rate only slightly above chance) in this condition resulted in
only slightly more than half of viewers' responses being included in the analysis of
the response data. This observation, coupled with the only slight differences in
latency for inverted same presentations, suggest that for inverted presentations in
Experiment 2, accuracy may be the best reflection of participant responding.
The superior accuracy observed for part faces upon such presentations
constitutes strong proof for out- daim that performance for inverted faces may be
mediated by part-based representational strategies associated with the object-
recognition system (Moscovitch et al., 1997); such strategies may result in attention
being focused upon the intemal features alone. Because these features may carry
the burden of information in facial recognition (Moscovitch et al., 1997), it is likely
that attention to these features is sufficient for recognition. In fact, the extemal
features may prove an unnecessary distraction at test when performance is driven
by feature-based cornparisons that are most readily available via the interna1
features.
Thus, although the task demands of Experiment 2 appeared to dictate more
global responding to upnght stimuli than was true in Experiment 1, we nonetheless
observed differential processing as a consequence of part versus whole
presentations of inverted comparison faces. That no evidence of superior
performance was found for inverted presentations of comparison faces comprised
of both the internal and external features strongly suggested that the extemal
features do not play a role in the reintegration or nonnalization of facial stimuli
following inversion.
Although the results of two previous studies (de Haan & Hay, 1986; Nachson
et al., 1995) suggest that the external features are more efficacious than the
intemal features in forming discrimination judgments for unfamiliar faces, the
findings of several other researchers (Ellis et al., 1979; Hines et al., 1987, Young et
al., 1985) are contradictory in that they found no reliable difference in identification
rates for unfamiliar faces when participants were presented with the internal or
external facial features. Nonetheless, participants in these studies were required to
make discrimination judgments of upright faces only. The results of the present
study suggest that part-based decornpositional strategies may result in a reliance
upon the internal rather than external features to form a discrimination judgrnent for
unfamiliar faces following inversion. Moreover, participants in both Experirnents 1
and 2 exhibited superior performance for presentation of both the internal and
external facial features upon upright same presentations of comparison faces,
suggesting that the extemal features aid in recognition performance for upright
same presentations of comparison faces. It is worth noting, however, that such
performance may be related to holistic processing strategies dictating attention to
the entire face, a factor seldom considered in determining the efficacy of the
internal and external features in forming discrimination judgments.
Experiments 3 and 4: Initial Presenta tion of Interna1 Fea tures
Experiment 3: Simultaneous Discrimination Task lnvolving Initial Presentation of
lntemal Features
Introduction. In a second set of experirnents, we sought to examine whether initial
presentation of the internal features only (part face) would heighten reliance upon
these features for discrimination. We reasoned that by focusing viewers' attention
on the internal features at study, it was likely that attention would continue to be
focused on these features at test.
In the present series of experirnents, initial encoding of the intemal features
was expected to result in superior performance being observed for part rather than
whole comparison faces, regardless of the orientation of test stimuli or the mode of
responding required. Specifically, although initial study presentations may still
have relied upon holistic encoding of an upright face, such faces were comprised
of the intemal features alone. Hence, initial viewer representations subsurned
these features only. Upon upright same presentations, holistic encoding of the
internal features at study was expected to focus attention on these features at test
(part face). By contrast, presentations of whole same faces were expected to result
in holistic encoding of the entire comparison face; such representations would be
incongruent with viewers' initial representations of the internal features alone.
Because upright different and al1 inverted presentations may rely upon part-
based encoding strategies, we again expected to observed superior performance
for part, rather than whole, presentations of corn parison faces. Indeed, evidence
accurnulated in Experiments 1 and 2 led us to betieve that the external features
may prove distracting at test when attention is focused upon the internal features
alone. Because part-based decompositional strategies require attention to
individual facial features such as the eyes, nose, and mouth, it is apparent that the
internal features which contain such information, may be the focus of viewer
attention when such strategies are required, thereby heightening the perceptual
saliency of these features. We did not expect, however, to observe as great of a
magnitude of difference in responding to part versus whole cornparison faces upon
such presentations as was expected for upright same presentations of comparison
faces. Although initial representations of the intemal features alone at study may
require that viewers "disentangle" these features from the whole upon cornparison
presentations of an entire face, this process would be far less difficult than for
upright same presentations that result in holistic or blended, rather than separate
representations, of the internal and extemal features.
Participants. The participants were 24 mernbers (mean age = 21 -96 years;
rnean years of education = 16.78) of the University of Toronto at Mississauga
community; nine were male and 15 female. Al1 had either normal or corrected-to-
normal vision. Participants received either nominal payrnent or academic credit for
their participation in the experiment, which took approximately 1 5 minutes.
Amaratus. The apparatus was identical to that in Experiments 1 and 2.
Stimuli. The stimuli were the same as those in Experiments 1 and 2. The
present experiment differed from Experiment 1 and 2, however, in that study faces
consisted of the intemal features only. Hence, on each experimental trial,
participants were required to make a same-different judgment of an upright test
face (internal features only) and one of four possible comparison faces: i) upright
faces (whole face); ii) upright faces (intemal features only); iii) inverted faces
(whole face) and; iv) inverted faces (internal features only). Whereas study faces
were always presented upright, comparison faces consisting of either the entire
face or internal features alone were presented either upright or inverted.
Procedure. The procedure was identical to that in Experiment 1, with the
exception that study faces consisted of the internal featu res only. Stimuli were
presented centered on screen, with study and comparison faces vertically aligned.
Study faces, consisting of the interna1 features only, were presented at the top of
the screen; comparison faces consisting of either the entire face or the intemal
features alone were presented either upright or inverted at the bottom of the
screen.
Participants initially completed sixteen practice trials (eight same, eight
different), during which they were allowed to ask questions about the experimental
procedure. The actual test session consisted of 128 trials; siw-four trials were
same trials, the remaining 64 were different trials. Test trials were presented in a
different random order for each participant.
Results.
Accuracv. Inversion of test faces again resulted in decrements in viewer
performance. When these data were analyzed by same versus different
responding (see Table 5), superior performance was again observed for upright
presentations of part comparison faces.
-----Insert Table 5 about here-----
These results were confirmed using a factorial ANOVA design. Means were
entersd into a 2 (Mode of Responding: Same and Different) X 2 (Target Orientation:
Upright and Inverted) X 2 (Target Format: Part and Whole) factoria! ANOVA with
Mode of Responding, Target Orientation, and Target Format treated as within-
subjects factors. The results of testing revealed main effects of Target Orientation
(E(1, 23) = 51.03, g == .O01 , Target Format (E(1, 23) = 8.97, g < .01, and Mode of
Responding (E(1,23) = 12.00, g < .01. Thus, upright faces were processed more
accurately than were inverted faces and part faces were processed more
accurately than were whole faces. In addition, responses to same items were more
accurate than those to different items. A two-way interaction between Mode of
Responding and Target Format was also observed, (E (1,23) = 9.06, < .01); a
simple effects analysis revealed performance for part faces varied more as a
function of same versus different modes of rssponding (F: (1, 23) = 65.69, g c -001)
than did performance for whole faces (E (1,23) = 14.81, g c .001). The two-way
interaction between Mode of Responding and Target Format was not significant (E
(1,23) = 2.59, g > .05) nor was that between Target Orientation and Target Format
(E (1,23) = .07, Q > .05). A three-way interaction between Mode of Responding,
Target Format, and Target Orientation was not observed, (E (1, 23) = 1.80, g > .05).
Post-hoc comparison contrasts (REGW-Q) between upright and inverted
faces confimed that discrimination performance for faces presented upright was
better than that for faces presented inverted (g < .05); viewers also exhibited
superior performance for part rather than whole presentations of comparison faces
(REGW-Q, e .05) and for same rather than different responses (REGW-O, g c .05).
Latencv. Similar to the accuracy data, response times were faster for faces
presented upright than for faces presented inverted. Superior performance was
also obseived for upright presentations of part rather than whole faces, regardless
of whether a same or different response was required (see Table 6). Similarly,
presentation of parf faces resulted in superior performance for inverted faces, but
for same responses only.
These results were confirmed using a 2 (Mode of Responding: Same and
Different) X 2 (Target Orientation: Upright and Inverted) X 2 (Target Format: Part
and Whole) factorial ANOVA with Mode of Responding, Target Orientation, and
Target Format treated as within-subjects factors. As in Experiments 1 and 2,
medians were calculated for each subject for each condition; only reaction times for
accurate responses were included in the analysis. In addition, we eliminated from
the analysis al1 reaction times that fell more than two standard deviations from the
mean and calculated new medians using the remaining data. This procedure
resulted in elimination of less than 3.7 percent of the total observations. This
procedure resulted in elirnination of less than 3.8 percent of the total observations.
Testing revealed a main effect of Target Orientation (E(1, 23) = 31.47, p c .001);
upright faces were processed more efficiently than were inverted faces. There was
also a main effect of Target Format (E(1, 23) = 26.70, Q c .001); part faces were
processed more efficiently than whole faces. These results were confirmed by
post-hoc analysis using the REGW-Q (g e .05).
A two-way interaction was observed between Mode of Presentation and
Target Format (E (1, 23) = 6.85, g c .OS); differences between modes of responding
(same versus different) were more evident for comparison presentations of part (E
(1, 23) = 18.34, g < ,001) rather than whole (E (1,23) = .34, > .05) faces. The
three-way interaction between Mode of Responding, Target Orientation, and Target
Format also approached significance (E (1, 23) = 3.70, Q c .07); although it was
readily apparent that viewers exhibited superior performance for part, rather than
whole, presentations of comparison faces upon upright same (E (1, 23) = 37.87, c
.001) and different (E (1, 23) = 30.77, e c .001) presentations, as well as inverted
same presentations (E (1, 23) = 28.24, g c .001), such differences were not as
apparent following inverted different trials (E (1, 23) = 0.74, Q > .05).
Main effects of Mode of Responding (E (1, 23) = 1.75, g z .05) were not
significant nor were there any interactions between Mode of Presentation and
Target Orientation (E (1, 23) = 1.56, g > .05) or between Target Orientation and
Target Format (E (1, 23) = 2.30, Q > .05).
Discussion. In Expet-iment 3, superior performance was observed for upright,
rather than inverted, presentations of comparison faces. Moreover, the results of
the accuracy data indicated that viewers exhibited superior performance for part
than for whole presentations of comparison faces. Responding was faster to part
than to whole comparison faces upon al1 upright trials, as well as for inverted same
responses. These results confimed Our earlier hypothesis that initial presentation
of the interna1 features alone would result in superior performance being observed
for part rather than whole comparison faces. Although faster responding for part
than for whole comparison faces was not observed upon different inverted
presentations, an inspection of the response data indicates that this non significant
finding may stem from the greater amount of time viewers required to make a
judgment of a part face on different trials (as cornpared to inverted same trials);
different presentations may have encouraged viewers to attend more closely to
individual facial features than was true upon same presentations (e.g., Matthews,
1978; serial processing for different faces only).
As we had predicted, the greatest magnitude of difference in responding to
part rather than whole comparison faces occurred for upright same presentations
[approximately 690 milliseconds as cornpared to 572 (upright different), 597
(inverted same), and 96 (inverted different) milliseconds]. Holistic processing
strategies for upright same faces likely resulted in global encoding of the entire
face upon comparison presentations of the entire face, requiring viewers to
disentangle the interna1 features from their holistic representation upon such trials.
Although such parsing would be required upon upright different and inverted
presentations of whole faces as well, part-based encoding strategies likely
resulted in separate representations of the intemal and external facial features,
making this process easier to accomplish. Moreover, part-based processing
strategies focusing attention upon the intemal features alone may have heightened
the perceptual saliency of these features upon such judgments. Although overall
responding to part and whole faces was slower for upright different and al1 inverted
presentations than it was for upright same faces, it was likely that the slower
response times observed for these stimuli can be attributed to the piecemeal nature
of responding to these items, when compared to the more rapid or global nature of
responding to upright same items.
Indeed, the results of the present experiment lend confirmation to our earlier
hypothesis that the extemal features rnay prove disruptive at test when
discrimination judgments can be fonned based upon the internal features alone.
The slower and less accurate responding observed for presentations of
comparison faces comprised of both the internal and external features for al1
upright presentations, as well as inverted sarne responses, indicates that these
features add no new or informative information for discrimination. Rather, it
appears that these featu res rnay prove disruptive at test when attention rernains
focused on the internal features. Indeed, these findings allow us to more strongly
reject our hypothesis that the external features serve as an anchoring point for
reintegration or norrnalization of facial stimuli following inversion; performance was
worse when the extemal features were present than when they were absent.
Experiment 4: Sequential Discrimination Task lnvolving Initial Presentation of the
Intemal Features
Introduction.
Because the third experiment involved simultaneous presentations of test
and comparison items that may have allowed for direct feature-by-feature
cornparisons to occur, we reasoned that a sequential paradigm like that used in
Experiment 2 would preclude the possibility that participants could rely upon this
mode of responding to facilitate discrimination. Hence, the procedure used in
Experiment 4 was identical to that in Experiment 2, with the exception that at study
faces consisted of the intemal features only. We reasoned that this paradigm may
result in more global encoding of test stimuli, at least upon initial study
presentations, than may have been true in Experiment 3, where participants would
have more opportunity to rely upon feature-by-featu re corn parisons than was
possible under the present experimental paradigm. Indeed, we expected that the
lack of difference in response times for part versus whole faces observed upon
inverted different presentations in Experiment 3 might be eliminated under the
present paradigm which required participants to retain a holistic representation of
the internal features presented at study and make judgments based upon their
mernory representations of the stimuli.
Participants. The participants were 24 members (mean age = 20.88; mean
years of education = 1 6.79) of the University of Toronto at Mississauga community;
seven were male and 17 female. AI1 had either normal or corrected-to-normal
vision. Participants received either nominal payment or academic credit for their
participation in the experirnent, which took approximately 15 minutes.
Apparatus. The apparatus was identical to that in the preceding
experiments.
Stimuli. The stimuli were identical to those in Experiment 2, with the
exception that comparison faces consisted of the internal features of the face only.
Procedure. The procedure was identical to that of Experiment 2, with the
exception that study faces consisted of the internal features only.
Results.
Accuracv. Inversion of test faces again resulted in decrements in viewer
performance. When these data were analyzed by same versus different modes of
responding, superior performance was again observed for upright and inverted
presentations of part comparison faces (see Table 7).
---4 nsert Table 7 about here-----
These results were confimed using a factorial ANOVA design. Means were
entered into a 2 (Mode of Responding: Same and Different) X 2 (Target Orientation:
Upright and Inverted) X 2 (Target Format: Part and Whole) factorial ANOVA with
Mode of Responding, Target Orientation, and Target Format treated as within-
subjects factors. The results of testing revealed main effects of Target Orientation
(E(l,23) = 37.88, Q < ,001, Target Format (E(1, 23) = 24.70, g .001, and Mode of
Responding (E(1.23) = 21.67, p c .OUI. Thus, upright faces were processed more
accurately than were inverted faces and part faces were processed more
accurately than were whole faces. In addition, responses to sarne items were more
accurate than those to different items. A two-way interaction was also observed
between Mode of Responding and Target Orientation, (E (1,23) = 9.29, g c .01);
planned contrasts revealed performance varied more as a function of same versus
different modes of responding (E (1, 23) = 56.41, g < .001) than did performance for
inverted faces (E (1, 23) = 10.24, g < .01). The three-way interaction between Mode
of Responding, Target Format, and Target Orientation (E (1, 23) = 4.17, g > .06) was
marginally significant. Planned comparisons revealed that performance accuracy
less as a function of part versus whole presentations for upright same comparison
faces (E (1, 23) = 3.72, g >.05), than for upright different, (E (1, 23) = 30.26, g c
.001) and inverted same (E (1,23) = 17.44, Q .001) and different (E (1, 23) =
13.44, g < .01) faces. The two-way interaction between Mode of Responding and
Target Format was not significant (E (1, 23) = .43, g > .05) nor was that between
Target Orientation and Target Format (E (1, 23) = .03, g > .05).
Post-hoc comparison contrasts (REGW-Q) between upright and inverted
faces confirrned that discrimination performance for faces presented upright was
better than that for faces presented inverted (e c .05); viewers also exhibited
superior performance for part than for whole presentations of cornparison faces
(REGW-Q, g < .05) and for same rather than different responses (REGW-Q, e c .05).
Latencv. Response times were faster for faces presented upright than for
faces presented inverted. More interestingly, however, responding was faster to
part rather than whole presentations of comparison faces, regardless of mode of
responding (same versus different) required or the target orientation (upright
versus inverted) of stimuli (see Table 8).
-----ln sert Table 8 about here-----
These results were confirmed using a 2 (Mode of Responding: Same and
Different) X 2 (Target Orientation: Upright and Inverted) X 2 (Target Format: Part
and Whole) factorial ANOVA with Mode of Responding, Target Orientation, and
Target Format treated as within-subjects factors. As in Experiments 1 and 2,
medians were calculated for each subject for each condition; only reaction times for
accurate responses were included in the analysis. In addition, we eliminated from
the analysis al1 reaction tirnes that fell more than two standard deviations from the
rnean and calcuiated new rnedians using the remaining data. This procedure
resulted in elimination of less than 3.6 percent of the total observations. Testing
revealed a main effect of Target Orientation (E(1, 23) = 24.74, c .001); upright
faces were processed more efficiently than were inverted faces. There was also a
main effect of Target Format (E(1, 23) = 27.28, g c .001); part faces were processed
more efficiently than were whole faces. A main effect of mode of responding (E(1,
23) = 8.65, g c .Ol) revealed that same presentations resulted in faster responding
than was observed for different responses. These results were confined by post-
hoc analysis using the REGW-Q (g e .05).
A two-way interaction was observed between Mode of Presentation and
Target Orientation (E (1, 23) = 6.85, g c .05); differences between modes of
responding (same versus different) were more evident for comparison
presentations of upright than inverted faces. In contrast to Experiment 3, however,
the three-way interaction between Mode of Responding, Target Orientation, and
Target Fonat was not significant (E (1, 23) = .23, g > .05); response performance
was better for presentation of part rather than whole comparison faces regardless
of mode of responding and target orientation. There were no interactions between
Mode of Presentation and Target Format (E (1, 23) = .33, g > .05) or between Target
Orientation and Target Format (E (1 , 23) = 1.43, g > .05).
Discussion. The results of Experiment 4 were relatively clear in suggesting that
initial attention to the internal features alone heightened viewers' reliance upon
these features at test. Similar to the results reported in Experiment 3, viewers
exhibited faster and more accurate performance for part rather than whole
presentations of comparison faces. Whereas in Experiment 3, however,
differences in response time were not apparent for inverted different cornparison
faces, the present experimental paradigm appeared to heighten viewers' reliance
upon holistic encoding at least upon initial presentation of an upright study face,
resulting in superior performance being obsetved for part than for whole
comparison faces, regardless of the orientation of test faces, or the mode of
responding required.
We once again observed the greatest differences in response time to part
versus whole faces upon upright presentations of sarne comparison faces.
Although the three-way interaction between Target Orientation, Target Format, and
Mode of Responding did not prove to be significant, performance for upright same
comparison faces varied by approximately 196 msec for part versus whole
comparison faces, as compared to 150 msec for upright different responses and
11 9 and 75 msec for inverted same and different presentations, respectively.
Hence, viewers appeared to have the greatest difficulty disentangling the internal
features from holistic representations of the entire face encouraged by global
responding to upright same presentations. Although overall responding to part and
whole faces was slower for those comparison faces that appeared to rely upon
part-based processing strategies (i.e., upright different and al1 inverted
presentations) than it was for upright same faces, it was likely that the slower
response times observed can be attributed to the piecemeal nature of responding
to these items, when compared to more rapid or global responding to upright same
items. Moreover, although greater differences in accuracy were observed for part
versus whole presentations for upright different and al1 inverted comparison
presentations than for upright same responses, the near-ceiling accuracy in this
condition (approximately 94 percent for part faces and 90 percent for whole faces)
may have precluded the emergence of any substantial differences in accuracy for
upright same faces.
Part-based encoding strategies once again appeared to result in a focus of
attention upon the internal features; superior performance was observed for part
rather than whole presentations of comparison faces. Indeed, initial attention to the
internal features alone rnay have rendered the external features distracting at test,
such that superior performance was observed when these features were absent
rather than present. Although viewers' initial holistic representation of a part face at
study rnay have required them to disentangle the internal features from the whole
face upon upright same presentations of the internal and extemal features, the
heightened perceptual saliency of the internal features, as well as separate
representations of the internal and external features that likely occurred as the
result of part-based decompositional processing strategies, for upright different and
al1 inverted presentations may have facilitated this process for these presentations.
Although recent evidence accumulated by Moscovitch and his colleagues
(Moscovitch et al., 1997) with the patient CK suggests that his difficulty in
recognizing inverted (as well as fractured) faces may have arisen from an inability
to appreciate the relation of parts of these stimuli to the whole (i.e., reintegration of
the face to a perceptual gestalt), the results of the present series of experiments are
consistent with those of Experiments 1 and 2 in suggesting that successful
recognition performance following inversion is possible by attending to the internal
features alone. Moreover, whereas viewers in Experiment 1 and 2 exhibited
superior performance for upright same presentations of whole faces comprised of
both the intemal and external features, viewers in the second series of experiments
were better able to form their discrimination judgments in this condition based on
consideration of the internal features alone. It is likely that initial attention to these
features alone upon upright study presentations heightened later reliance upon the
intemal features for discrimination. Hence, although two studies (de Haan & Hay,
1986; Nachson et al., 1995) that used different experimental methods provided
evidence that external features are more efficacious than internal features in
recognition tasks for upright unfamiliar faces, the results of the present series of
experiments suggest that these features rnay prove distracting at test when
attention is focused on the intemal features alone at study.
General Discussion
The central issue addressed in this study was whether differences in
processing strategies for upright and inverted faces (Farah et al., 1995; Moscovitch
et al., 1 997; Tanaka & Farah, 1 993), as well as in same versus different modes of
responding to upright faces (Bradshaw & Wallace, 1971; Matthews, 1978; Smith &
Neilson, 1970; Sergent, 1984) would mediate responding to upright and inverted
comparison faces comprised of either the internal and external features orthe
internal features alone. An additional aim of the present study was to determine
whether the internal and external features played a differential role in
discrimination performance upon upright and inverted presentations. Our results
provide new evidence that whereas holistic processing strategies mediate
responding to upright same facial stimuli, more analytic or part-based
decompositional strategies rnay be responsible for the processing of upright
different and al1 inverted facial stimuli. Moreover, attending to the internal features
alone was sufficient to form a discrimination judgment for upright different and al1
inverted faces; the same was true for upright same faces when attention was
focused on the internal features alone ai study. By contrast, viewers relied upon
both the internal and external features to form a discrimination judgment for upright
same faces when attention was focused on the entire face at study.
We found no evidence that the external features seive as an anchoring point
for reintegration or normalization of facial stimuli following inversion. Rather, the
superior performance observed upon inverted presentations in Experiment 2 and
across al1 presentations in Experirnents 3 (with the exception of inverted different
presentations) and 4, when the external features were absent, rather than present,
suggests that these features rnay prove disruptive at test when attention is focused
upon the internal features alone.
Our findings were consistent with the suggestion that different processing
strategies may rnediate responding to same versus different presentations of
upright faces (Bradshaw & Wallace, 1971 ; Matthews, 1978; Smith & Neilson, 1970;
Sergent, 1984). Whereas two early studies (Bradshaw & Wallace, 1971 ; Smith &
Neilson, 1970) provided some preliminary evidence that responding to upright
different items at short retention intervals (simuitaneous presentations, as well as
delays of up to 4 sec between presentations of test and comparison faces) relies
predominately upon feature-by-feature processing strategies, the results of later
studies (Matthews, 1978; Sergent, 1984) were contradictory in that they
demonstrated that responding to these stimuli rnay rely upon both holistic and part-
based encoding strategies. The results of the current investigation indicate that
whereas responding to upright sarne faces may rely upon holistic or configurational
processing strategies dictating attention to the entire face, responding to upright
different items may instead rely upon more feature-based or analytic processing
mechanisms dictating attention to the internal features alone, which appear to carry
the burden of information processing in facial recognition (Moscovitch et al., 1997)
When initial study presentations involved whole faces, performance was better for
whole than for part presentations of upright same comparison faces. By contrast,
initial encoding of a whole face at study appeared to overwhelm more part-based
representations of upright different comparison faces at test, resulting in no
differences being obsewed in performance for upright different presentations in
Experiments 1 and 2.
When initial study presentations involved the intemal features alone,
however, performance upon upright same presentations was best for part rather
than whole presentations of comparison faces; difficulties in disentangling the
intemal features from holistic representations of whole comparison faces may
account for this finding. Performance for upright different responses was also
better for part than for whole presentations of comparison faces, although the
magnitude of difference between part and whole presentations of cornparison
faces was not as great as for upright same presentations, suggesting that separate
representation of the intemal and external features, as well as the heightened
salience of the intemal features as a result of part-based encoding strategies, may
have made it easier for viewers to disentangle the part from the whole upon such
presentations.
It is worth noting that, similar to the Matthews (1978) and Bradshaw and
Wallace (1 971 ) investigations, our experiments involved either simultaneous
presentations of study and comparison stimuli, or brief one-second delays between
study and test presentations. It is unclear whether differences in responding to
same and different items would emerge at longer retention intervals that may rely
more upon long-term or consolidated memory representations. Nonetheless,
although processing of upright faces appears to rely predominately upon holistic
encoding strategies described by other authors (e.g., Farah et al., 1995; Yin, 1969),
it is clear that facial-recognition system is able to accommodate the part-based
representations that rnay be required for processing of different faces. Whether
such representations rely upon serial processing mechanisms, however, while
responding to same items relies upon more parallel processing mechanisms is one
interesting area awaiting further investigation.
We speculate that part-based decompositional strategies encouraged by the
object-recognition systern that appears to be responsible for the processing of
inverted faces (Farah et al., 1995; Moscovitch et al., 1997) may have dictated a
focus of attention on the internal features alone at inversion, resulting in superior
performance being observed for part rather than whole presentations of
cornparison faces in Experiments 3 (with the exception of different responses) and
4; viewers also exhibited superior performance for part faces at inversion in
Experiment 2. No evidence was found that the external features, which appear to
have preferential access to the object-recognition system following inversion
(Moscovitch et al., 1997), aided in the perceptual reintegration or nonalization of
inverted stimuli. Because it is apparent that the intemal features carry the burden
of information in facial recognition processing, i t appears that attention to these
features alone is sufficient to form a discrimination judgment under conditions of
inversion. Moreover, because part-based decompositional strategies require
attention to individual facial features such as the eyes, nose, and mouth, it is
apparent that the internal features, which contain such features, are the likely focus
of attention when such strategies are required.
Because part-based decompositional strategies may focus attention on the
internal features at test, as well as facilitate separate representations of the intemal
and external features following comparison presentations of whole faces, we did
not observe as great of differences in responding to part versus whole inverted
faces as we did for upright same items in Experiments 3 and 4. We speculate that
the heightened perceptual saliency of the intemal features, as well as the
individual consideration allowed the internal and external features due to part-
based encoding strategies, may have made it easier to disentangle the intemal
features upon comparison presentations of whole inverted faces. Indeed, the
external features may prove distracting at test under such conditions; we also
observed interference by the external features upon both upright same and
different presentations in Experiments 3 and 4, suggesting that when attention is
focused primarily upon the intemal features alone these features may prove
distracting, regardless of the orientation of the stimuli.
Although the results of two previous studies (de Haan & Hay, 1986; Nachson
et al., 1995) suggest that the external features are more efficacious than the
internal features in forming discrimination judgments for unfamiliar faces, the
findings of other researchers (Ellis et al., 1979; Hines et al., 1987, Young et al.,
1985) are contradictory in that they found no reiiable difference in identification
rates for unfamiliar faces when participants were presented with the intemal or
external facial features. Nonetheless, participants in these studies were required to
make discrimination judgments of upright faces only. The results of the present
study make it clear that part-based decompositional strategies may result in a
reliance upon the internal rather than external features to form a discrimination
judgment for unfamiliar faces following inversion.
Participants in both Experiments 1 and 2 exhibited superior performance for
presentation of comparison faces consisting of both the intemal and external facial
features upon uprig ht sarne presentations, suggesting that the external features aid
in recognition performance for such unfamiliar faces. It is important to note,
however, that such performance may be related to holistic processing strategies
dictating attention to the entire face; no differences were observed in performance
for upright different comparison faces which may rely upon more part-based
processing mechanisms. When attention was focused on the intemal features at
study in Experiments 3 and 4, viewers exhibited superior performance for
comparison faces comprised of the intemal features alone upon both same and
different comparison presentations of upnght faces. The short conclusion is that
whereas the external features may aid in recognition performance for upright faces
when attention is focused upon the entire face and holistic processing strategies
are required (as for upright same judgments), the intemal features alone may be
sufficient to f o n a discrimination judgment when attention is focused upon these
features alone, in particular when part-based encoding strategies are required.
One alternative explanation for our findings is that format congruency
between test and comparison faces contributed to superior performance being
observed for comparison faces that shared the same features as those presented
at test (e.g., part face to part face). A recent series of experiments by Harman and
Moscovitch (in preparation) both replicated and extended the findings of other
researchers (e.g., Bruce, 1982; 1988; Farah et al., 1995; Patterson & Baddeley,
1977) in suggesting that format incongruency (i.e., differences in orientation or
form) between test and studied items may lead to subsequent detriments in
recognition performance. In Harman and Moscovitch's (in preparation) experiment,
participants were required to make an old-new judgment of faces and objects
presented either upright or inverted orfractured and whole. Half of the stimuli were
incongruent at study and test (e.g., whole to inverted) and the remaining half were
congruent (e.g., inverted to inverted). If congruence between study and test items
facilitates recognition, then items tested in a congruent format should exhibit
superior performance, regardless of whether they are presented upright or inverted
(or fractured or whole). Indeed, the results of the experiment indicated that facial-
recognition performance was best when the orientation of items at study matched
that at test, regardless of whether items presented were upright or inverted (or
fractured or whole). Recognition of objects, however, was best when test stimuli
were presented upright, regardless of study orientation.
In the present experiment, viewers exhibited superior performance for
comparison faces sharing the same facial features as those at test (e.g., part face to
part face). Nonetheless, such results were rnediated by processing strategies
associated with the recognition of upright (same and different) and inverted faces.
Specifically, viewers in Experiments 1 and 2 were presented with whole faces at
study; superior performance for whole faces was observed only upon upright same
presentations of cornparison faces that likely relied upon holistic or configurational
processing mechanisms. Indeed, superior performance was observed for part
faces following inverted comparison presentations in Experiment 2 perhaps
reflecting the contribution of part-based processing mechanisms mediated by the
object-recognition system. Study presentations in Experiments 3 and 4 involved
the intemal features only. With the exception of inverted different trials in
Experiment 3, superior performance was observed for part (interna1 features only)
rather than whole comparison trials, regardless of the orientation of comparison
stimuli or the mode of responding required; 2 reliance upon either holistic (upright
same faces) or part-based (upright different and inverted faces) processing
strategies could account for these findings. Because processing strategies (i.e.,
holistic or part-based) appear to supersede format congruencies between test and
cornparison items in detemining viewer performance, we suggest that a reliance
upon differential processing strategies for upright (same and different) and inverted
faces provides a better explanation of the current findings than does format
congruency.
Performance in several experimental conditions was relatively low. In
particular, viewers found it difficult to discriminate sequentially-presented inverted
cornparison faces, limiting the number of latency responses that were included in
the analysis of these conditions. Future research may incorporate testing
paradigms that provide for more accurate responding by viewers (e.g., longer
presentations of facial stimuli). In addition, although previous research suggests
that the object-recognition systern has preferential access to the external features
(Moscovitch et al., 1997) under conditions of inversion, the results of the present
study suggest that viewers may rely primarily upon the interna1 features to rnake a
discrimination judgment. Nonetheless, viewers were instructed to explicitly attend
to these features in forming their discrimination judgments. Future experimental
paradig ms rnay inco rporate different external features at study and test
presentations to determine more conclusively the role of the extemal features in
discrimination performance. In addition, other experiments, involving inverted
presentations of study faces may also be required to assess more carefully the
effects of part-based processing strategies ai initial study. Nonetheless, the results
of the present series of experirnents make it clear that viewers appear to rely upon
differential processing strategies in the recognition of upright (same and different)
and inverted faces. Moreover, in the present study attending to the intemal
features alone was sufficient to form a discrimination judgment for upright different
and al1 inverted faces; the same was true for upright same faces when attention
was focused on the intemal features alone at study. By contrast, viewers relied
upon both the intemal and external features to form a discrimination judgment for
upright same faces when attention was focused on the entire face at study.
References
Allison, T., Ginter, H., McCarthy, G., Nobre, A. A., Puce, A., Luby, M., &
Spencer, D. D. (1994). Face recognition in human extrastriate cortex. Journal of
Neuroph~sioloav. 71.821 -825.
Bagnara, S., Boles, D. B., Simion, F., & Umiltà, C. (1 982). Can an analytic/
holistic dichotomy explain hemispheric asymmetries? Cortex,J& 67-78.
Banich, M. T. (1 997). Neuropsvcholoav: The Neural Basis of Mental
Funciion, Chapter 5, Object Recognition. New York: Houghton Mifflin, pp. 234-273.
Bartlett, J. C., & Searcy, J. (1993). Inversion and configuration of faces.
Coanitive Psvcholoav. 25,281 -31 6.
Bentin, S., Allison, T., Puce, A., & McCarthy, G. (1 996). Electrophysiological
studies of face perception in humans. Journal of Coanitive Neuroscience. 8,551 -
565.
Behmann, M., Moscovitch, M., & Winocur, G. (1 994). Intact visual imagery
and impaired visual perception in a patient with visual agnosia. Journal of
Experimental Psvcholoav: Human Perception and Performance. 30, 1068-1 087.
Behmann, M., Moscovitch, M., & Winocur, G. (1 992). Dissociations
between mental imagery and object recognition in a brain-damaged patient.
Nature. 359,636-637.
Biederman, I., & Gerhardstein, P. C. (1 993). Recognizing depth-rotated
objects: Evidence and conditions for three-dimensional viewpoint invariance.
Journal of Experimental Psvcholoav: Human Perception and Performance, 19,
1 162-1 182.
Bradshaw, J. L., & Wallace, G. (1971). Models for the processing and
identification of faces. Perception and Psvchophvsics. 9.443-448.
Bruce, V. (1 982). Changing faces: Visual and nonvisual coding processes
in face recognition. British Journal of Psycholoav. 73, 105-1 16.
Bruce, V. (1 988). Recoanisina Faces. Erlbaum, UK.
Bruce, V., & Young, A. (1 986). Understanding face recognition. British
Journal of Psvcholoav. 77,305-327.
Campbell, R., & Tuck, M. (1 995). Recognition of parts of famous-face
photographs by children: an experimental note. Perception. 24.451-456.
Darnasio, A. R. (1990). Category-related recognition deficits as a clue to the
neural substrates of knowledge. Trends in Neuroscience. 13,95-98.
Davies, G., & Flin, R. (1 984). The man behind the maçk-disguise and face
recognition. Human Leaminu. 3,83-95.
Davies, G., Ellis, H. D., & Shepherd, J. W. (1 978). Face recognition
accuracy as a function of mode of representation. Journal of A~pl ied Psvcholoav,
63. 180-1 87.
Diamond, R., & Carey, S. (1986). Why faces are and are not special: An
effect of expertise. Journal of Experirnental Psvcholoav: General. 1 15, 107-1 17
de Haan, E. H. F., & Hay, D. C. (1 986). The matching of famous and
unknown faces, given either the intemal or extemal features: A study of patients
with unilateral brain lesions. In H. D. Ellis, J. A. Jeeves, F. Newcombe, & A. Young
(Eds.). Aspects of face proceçsing (pp. 302-309). Dordrecht: Nijhoff.
de Haan, E. H. F., Young, A., & Newcombe, F. (1 987). Face recognition
without awareness. Coanitive Neuropsvchologv. 4, 385-41 5.
Desimone, R. (1 991 ). Face-selective cells in the temporal cortex of
monkeys. Journal of Coonitive Neuroscience. 3,1-8.
Ellis, H. D., Shepard, J. W., & Davies, G. M. (1 979). Identification of familiar
and unfamiliar faces from intemal and extemal features: Source implication for
theories of recognition. Perception, 8,431 -439.
Eriksen, C. W., O'Hara, W. P., & Eriksen, B. (1 982). Response competition
effects in same-different judgrnents. Perception and Psvchophvsics. 32,261 -270.
Esgate, A., Burton, A., & Burton, E. (1996). Lateral advantages in same-
different comparisons of two-note, dichotically presented chords to a successively-
presented probe. Neuropsvcholoaia. 35, 1 77-1 82.
Fagan, J. F. 111 (1 973). Infants' delayed recognition, memory, and forgetting.
Journal of Experimental Child Psvcholoav. 16.424-450.
Farah, M. J. (1 990). Visual aanosia: Disorders of obiect recoanition and
what thev tell us about normal vision. Cambridge, MA: MIT Press.
Farah, M. J., Tanaka, J. W., & Drain, H. M. (1995). What causes the face
inversion effect? Journal of Experimental Ps choloav: Human Perception and
Performance, 21.1-7.
Hanan, K., & Moscovitch, M. (in preparation). A re-evaluation of the
disproportionate face inversion effect.
Hawkins, H. L., McDonald, G. J., & Cox, A. K. (1973). Effects of irrelevant
information in speeded discrimination. Journal of Ex~erimental Psycholoav. 98,
435-437.
Haxby, J. V., Howitz, B., Ungerleider, L. G., Maisog, J. M., Pietrini, P., &
Grady, C. L. (1994). The functional organization of the human extrastriate cortex: A
PET rCBF study of selective attention to faces and locations. Journal of
Neuroscience. 14, 633696353.
Hines, D., Jordan-Brown, L., & Juzwin, K. R. (1 987). Hemispheric visual
processing in face recognition. Brain & Coanition. 6, 91-100.
Hock, H. S. (1 973). The effects of stimulus structure and familiarity on
same-different comparisons. Perception and Psvcho~hvsics. 14,413-420.
Horel, J. (1 993). Retrieval of a face discrimination during suppression of
monkey temporal cortex with cold. Neurops~choloaia. 31. 1067-1 077.
Humphreys, G. W., Riddoch, M. J., Donnely, N., Freeman, T. Boucart, M., &
Muller, H. M. (1994). Intermediate visual processing and visual agnosia. In M. J.
Farah & G. Ratcliff (Eds.). The neuropsvcholoav of hiah-level vision (pp. 63-1 01).
Hillsdale, NJ: Erlbaum.
Marr, D., & Nishihara, H. K. (1 978). Representation and recognition of the
spatial organization of three-dimensional shapes. Proceedings of the Roval
Societv of London. Series B. 200, 269-294.
Matthews, M. L. (1 978). Discrimination of Identikit construction of faces:
Evidence for a dual processing strategy. Perception and Psvcho~hvsics. 23, 153-
161.
Mauer, D. (1 983). The scanning of compound figures by young infants.
Journal of Ex~erimentai Child Psvcholoav. 35,437-448.
Mauer, D., & Salapatek, P. (1976). Developmental changes in the scanning
of faces in young infants. Child Development. 47,523-527.
Moscovitch, M., Behmann, M., & Winocur, G. (1997). What is special about
face recognition? Nineteen experiments on a person with visual object agnosia
and dyslexia but normal face recognition. Journal of Coanitive Neuroscience. 9,
555-604.
Nachson, I., Moscovitch, M., & Umiltà, C. (1 995). The contribution of
extemal and interna1 features to the matching of unfamiliar faces. Psvcholooical
Research, 58'31 -37.
Newcombe, Ç., Metha, Z., & de Haan, E. H. F. (1994). Category specificity in
visual recognition. In M. J. Farah & G. Ratcliff (Eds.). The neuropsvcholoav of hiah-
level vision (pp. 103-132). Hillsdale, NJ: Erlbaurn.
Patterson, K., & Baddeley, A. D. (1977). When face recognition fails.
Journal of Ex~erirnental Psvcholoav: Human Learninci and Memorv, 3,406-417.
Peters, D. P. (1 987). The impact of naturally occurring stress on children's
memory. In S. J. Ceci, M. P. Toglia, & D. F. Ross (Eds.). Children's evewitness
mernory (pp. 122-1 41). New York: Springer-Verlag.-
Püschel, J., & Zaidel, E. (1 994). The Benton-Van Allen faces: A lateralized
tachistoscopic study. Neuro~svcholoaia. 32, 357-367.
Ratcliff, R. (1 993). Methods for dealing with reaction time outliers.
Psvchological Bulletin. 1 14,510-532.
Rhodes, G., Brake, S., & Atkinson, A. P. (1993). What's lost in inverted
faces? Coqnition, 47'25-57.
Rock, 1. (1 973). Orientation and forrn. New York: Academic Press.
Searcy, J. H., & Bartlett, J. C. (1 996). Inversion and processing of
component and spatial relational information in faces. Journal of Experimental
Psvcholoav: Hurnan Perception and Performance. 23, 904-91 5.
Sergent, J. (1 984). An investigation into component and configurational
processes underlying face perception. The British Journal of Psycholo
242.
Sergent, J. (1 986). Microgenesis in face perception. In H.D. Ellis, J.A.
Jeeves, F. Newcombe, & A. Young (Eds.). Aspects of face processinq (pp. 17-73).
Dordrecht: Nijhoff.
Shapely, R., & Gordon, J. (1 985). Nonlinearity in the perception of form.
Perception & Psvchophvsics. 37,84088.
Shepherd, J. W., Ellis, H. De, & Davies, G. M. (1982). Identification evidence:
A psvcholoaical investiaat ion. Aberdeen: University of Abeerdeen Press.
Smith, E. E., & Neilson, G. D. (1 970). Representations and retrieval
processes in short-term memory: Recognition and recall of faces. Journal of
Experimental Psv-, 397-405.
Taylor, D. A. (1976a). Holistic and analytic processes in the comparison of
letters. Perce~tion and Psvchophvsics. 20, 187-1 90.
Taylor, D. A. (1976b). Effect of identity in the multi-letter matching task.
Journal of Experimental Psvcholoav: Human Perce~tion and Performance. 2. 117-
128.
Valentine, T. (1 988). Upside-down faces: A review of the effects of inversion
upon face recognition. British Journal of Psycholoay. 79,471 -491.
Valentine, T. (1991). A unified account of the effects of distinctiveness,
inversion, and race in face recognition. The Quarterlv Journal of Experimental
PSVC~O~OQV. 43, 161 -204.
Valenza, E., Simion, F. Cassia, V. M., and Urniltà, C. (1996). Face
preference at birth. Journal of Experimental Psvcholoav: Human Perception and
Performance, 22,892-903.
Vassilev, A., & Mitov, B. (1976). Perception time and spatial frequency.
Vision Research. 16, 89-92.
Yamane, S., Kaji, S., 8 Kwano, K. (1 988). What facial features activate face
neurons in the inferotemporal cortex of the monkey? Experirnental Brain Research,
73,209-214.
Yin, R. K. (1969). Looking at upside down faces. Journal of Experimental
Psvcholoav, 81, 141 -1 45.
Young, A. W. (1 984). Right cerebral hemisphere superiority for recognizing
the intemal and extemal features of famous faces. British Journal of Psvcholoa~
75. 161-169.
Young, A. W., Hay, D. C., McWeeny, K. H., Flude, B. M., & Ellis, A. W. (1985).
Matching familiar and unfarniliar faces on intemal and external features.
Perception. 14, 737-746.
Table 1
Ex~eriment 1: Mean Accuracv bv Mode of Response and Taraet Format
Upright l nverted
Mode of - M - SD - n - M - SD - n Response
Same
Target Format
Part
Whole
Different
Target Fo m a t
Part
Whole
Table 2
Ex~eriment 1 : Mean Reaction Time bv Mode of Response and Taraet F o n a t
Upright 1 nverted
Mode of - M - SD - n - M - SD - n Response
Same
Target F o n a t
Part
Whole
Different
Target F o n a t
Part 1582.15 303.49 24 2056.13 640.22 24
Whole 1645.04 479.69 24 2023.52 850.76 24
Table 3
Experiment 2: Mean Accurac bv Mode of Response and Target Format
Upright l nverted
Mode of Response
Same
Target Format
Part
Whole
Different
Target Format
Part
Whole
Table 4
Exueriment 2: Mean Reaction Time bv Mode of Response and Taraet Format
Upright l nverted
Mode of - M SD - n Response
Same
Target Format
Part
Whole
Different
Target Format
Part
Whole
Table 5
Experiment 3: Mean Accuracv bv Mode of Response and Taraet Format
Upright
Mode of Response
Same
Target Format
Part
Whole
Different
Target Format
Part
Whole
Table 6
Experiment 3: Mean Reaction Time bv Mode of Res~onse and Taraet Format
Mode of Response
Same
Target Format
Part 1203.77 339.36 24
Whole 1893.10 814.03 24
Oifferent
Target Format
Part 954.06 309.32 24 2351.04 1193.64 24
Whole 1037.41 358.1 0 24 2443.56 1161 -44 24
Table 7
Ex~eriment 4: Mean Accuracv bv Mode of Response and Taraet Format
Upright lnverted
Mode of - M SD - n - M - SD - n Response
.. - -
Same
Target Format
Part
Whole
Different
Targ et Format
Part
Whole
Table 8
Experiment 4: Mean Reaction Time bv Mode of Response and Target Format
Upright 1 nve rted
Mode of Response
Same
Target Format
Part
Whole
Different
Target Format
Part 1063.83 262.13 24 1143.90 271.73 24
Whole 1213.48 414.53 24 1218.46 358.64 24