Full text of DOI:10.3758/BF03196436

Memory & Cognition
2002, 30 (5), 811-821
The use of word–picture verification to study
entry-level object recognition:
Further support for view-invariant mechanisms
STEFANO A. DECARO
Fooyin University of Technology, Kaohsiung Hsien, Taiwan
and
ADAM REEVES
Northeastern University, Boston, Massachusetts
There aresubstantiallogicaland empiricalreasons for rejectingthe popular viewthatplane-misoriented
objects are identified afternormalizationof global orientation. Our subjectsdetermined the entry-level
identity of common objects (line drawings)
determining basic orientation (upright vs. rotated)—
before
as they must if they are to begin to know how to restore the image to the canonical upright. We used a
description–picture matching procedure in which 75% of the trials involved a mismatch in identity, ori-
entation, or both, so that comparisons could be based on one response ( ). Times to verify identity
no
were fasterthan times to verifyorientation and did not increasewith rotations in the picture plane. That
mismatch objects were positively identified at the entry level was shown both in a surprise recognition
test for object names and through the transfer of priming from the matching task to a subsequent object-
naming task. We conclude that classicmental rotation-like effectson naming times do not reflectearly
object encoding and recognitionprocesses,which are viewinvariant,but may stem from double-checking
at a postrecognition verificationstage.
The process of object recognition through vision in- of common objects showed that critical stimulus onset
evitablyconfoundsobject-specifyinginformation (such as asynchronieswere 2–19 msec shorter when identity,rather
surface, shape, color, and local features) with incidental than orientation, was verified and that it was orientation
information specifying the object’s orientation (typically, matching, not identity matching, that was most sensitive to
lighting and viewing angle). Thus, a critical question for rotations in the picture plane (DeCaro & Reeves, 2000).
theory is whether recognition processes can be brought to The response times (RTs) and error patterns from an ex-
bear directly on the relevant, object-specifying properties periment in which subjects matched pictures against ori-
upright car
or whetherincidentalinformation,such as orientation,must entation–name descriptions (such as
) also
be encoded first. The increase in recognition latency asso- showed a perceptual advantage for determining object
ro-
ciated with plane-rotated depictions of common objects identity over the simplest classification of orientation,
tated
upright
(e.g., Jolicœur, 1985) has been thoughtto imply that orien-
versus
(DeCaro, 1998). These results agree
tationinformation is extracted early and then used to trans- with, and extendto pictures, data from an early experiment
form the objectimage to match its canonical(upright)view by Corballis, Zbrodoff, Shetzer, and Butler (1978), in
prior to recognition.
which subjects were faster to verify the identity of al-
However, if recognition can be based on view-invariant phanumeric characters than to verify the characters’ an-
properties,successful encodingof objectclass may precede gular orientations.
the extraction of orientation. Psychometric data from a
Theories that place the process of object identification
backward-masking experiment which used line drawings before the process of orientation identification are natu-
rally compatible with view-invariant accounts of object
recognition, because early view-invariant object recogni-
tion cannot support the determination of orientation rela-
This article is based on a dissertation presented by the first author in
tive to an externalframe. Rather, the visual–cognitivesys-
partial fulfillment of the requirements for the Ph.D. degree at North-
tem must derive eithera viewer-centeredoran environment-
eastern University. Portions of this article were presented by the second
centered descriptionof the object, and this is likely to take
more time or require additionalprocesses, such as mental
rotation.Although view-invariantaccountsshould predict
that orientation is extracted after identity, findings to this
effect need to be supplementedby more direct support for
the use of view-invariant mechanisms in visual object
author at the first annual meeting of the Vision Sciences Society,
Ft. Lauderdale, FL. We thank Michael Corballis and Ruth Maki for their
careful reviews and encouraging comments on the original draft of this
article. Correspondence concerning this article should be addressed to
S. A. DeCaro, Department of Applied Languages, Fooyin University of
Technology, 151 Chinhsueh Road, Ta-Liao Hsiang, Kaohsiung Hsien
831, Taiwan (e-mail: sdecaro@lynx.dac.neu.edu).
811
Copyright 2002 Psychonomic Society, Inc.

E
D CARO AND REEVES
812
recognition. To this end, we present new experiments cases, by themselves, cannot indicate which element is
based on the description–depiction classification proce- verified first, but they can reveal whether performance in
dure introduced by DeCaro (1998), and we address im- the fastest partial-mismatchconditionis influencedby the
portant issues regarding its use.
second mismatch. Such effects are shown by an increase
in error rates and RTs to verify partial mismatches, rela-
tive to error rates and RTs to verify global mismatches—
Word–Picture Matching
The procedure we employed in this study is a variant of termed a
global mismatch effect
.
the word–picture matching task often used to probe recog-
The data reported by DeCaro (1998) showed clearly
nition processes. In the typical word–picture experiment, that (1) subjects were faster to detect a mismatch object
subjects view a word and then a picture in sequence, and than a mismatch orientation and (2) identity mismatches
yes
the task is to report whether the picture matches the word rarely elicited a false
response, whereas a significant
onthebasisof specifiedcriteria(e.g., Hamm & McMullen, number of orientation mismatch cases were incorrectly
description picture
an orientation–name description (such as
1998).In our
–
experiment,subjectsview classified as matches. Both findingspoint to a primacy for
UPRIGHT TREE
), extracting object class: Identity is encoded before orien-
followed by a line-drawing depiction of a common object tation, and when orientationis encoded,identityinforma-
(see Figure 1). The critical trials are those involvinga par- tion tends to dominatethe response. In this article, we first
tial mismatch, either in the depicted object’s name (an consider two potential processing models that might ex-
identity mismatch
) or in the depicted object’s orientation plain the original data: a
(an
termine identity and orientation is then inferred from the tive to the process of specifying identity, and a
shift model
, in which the process
orientationmismatch
). The relativetime needed to de- of specifying orientation is completely time-shifted rela-
cross
no
time taken to correctly report to each of these types of
model
, in which the process of specifying orientation be-
mismatches. Thus, the description–picture procedure is gins first but ends sufficiently later to give rise to overall
pairs
unconventional, because it uses
of words to probe slower mean RTs. These two processing models are dis-
two types of encodingon each trial and because the match tinguished in Experiment 1 with psychometric functions.
yes
(
) trials are eventually discarded and only the mis-
no
We also address a methodological issue concerning
specificity in matching tasks. Naming responses have the
match ( ) cases are used to measure performance.
A methodologicaladvantageof the description–picture advantageof specifyinga levelofclassification—typically,
basic level
procedure is that the critical trials are compared by using the
(Rosch, Mervis, Gray, Johnson, & Boyes-
the same response, task, and subject. In addition, these Braem, 1976), whichis also the entry pointfor most objects
comparisonsare supplementedby performance in thethird, (see Jolicœeur, Gluck, & Kosslyn, 1984). Although the
globalmismatch
condition,in which bothidentityand ori- descriptionsin our task are also at the basic or entry level,
entation can signal a mismatch. The global mismatch it is not clear whether the object depictions are classified
Figure 1. Overview of the description–picture matching procedure. The task is to report yes
if both the identity and the orientation of the depicted object match the preceding description
and to report no otherwise. This example illustrates an identity mismatch trial.

OBJECT IDENTIFICATION AND ORIENTATION
813
at this level of specificity, because global shape informa- replication serves primarily to facilitate our psychometric
tion alone may be sufficient to signalan identitymismatch. analysis and subsequent discussion of the original data.
In this study, we employed a recognition test (Experi-
ment 2) and an intertask priming procedure (Experi- Method
Subjects. Sixteen undergraduate students participated in Exper-
iment 1.
Stimuli and Equipment. The test displays were 96 black line
drawings of common objects, presented against a white background
ment 3B) to investigate the specificity of the verification
judgments. Finally, in Experiment 3A, we will present a
modified version of the description–picture procedure
that permitted a direct test of rotation effects on verifica-
on a 15-in. ViewSonic monitor with 640 3 480 pixel resolution. Re-
tion times.
sponses were recorded with a two-button serial mouse (Segalowitz
& Graves, 1990) attached to an IBM-compatible 80486 microcom-
puter. The computer controlled the output of the video display and
measured RTs through a software millisecond timer with 1-msec
precision (Graves & Bradley, 1987, 1991).
Procedure. The subjects held the computer mouse with both
hands, using their left and right thumbs to enter each response. Re-
sponse button assignments were set by each subject according to his
or her preference (since there was no need to counterbalance). Each
experimental trial began with the presentation of the description,
which appeared directly above a black ring centered on an otherwise
white display. (The ring subtended 10º of visual angle, which was
just large enough to contain the tallest and widest objects). The de-
scription specified an orientation, either the word UPRIGHT or the
GENERAL METHOD
All the subjects in this study were undergraduate students, who
participated in partial fulfillment of their psychology course re-
quirements at Northeastern University. To be eligible for participa-
tion, students were required to be native speakers of English and to
demonstrate at least 20/30 Snellen acuity at the time of the experi-
ments. The various description–picture sessions were always pre-
ceded by a block of 32 practice trials, during which simple alphanu-
meric characters were displayed and matched against preceding
descriptions, such as UPRIGHT NUMBER or ROTATED LETTER. In this
way, our subjects were sufficiently prepared for the experiments
without viewing any object depictions.
The testing items in this study were line drawings of common ob-
jects taken from the Snodgrass and Vanderwart (1980) picture set
(121 items) and the Microsoft ClipArt gallery (7 items). These de-
pictions were judged beforehand to have a standard upright view.
Sixteen students in a pilot study were asked to point to the “top” side
of each object once with a computer mouse. An object was consid-
ered to have a canonical view in the picture plane if all the students
selected the same side of the object, regardless of the object’s orien-
tation in the display.
Our methods in this study speak primarily to entry-level (rather
than basic-level) identification, because some of the word probes were
subordinate-level descriptions (e.g., rocking chair and wine glass).
We did not limit our descriptions to the basic level, because we were
concerned with early recognition and many objects are first identi-
fied (have their entry point) at the subordinate level because they are
atypical members of basic-level categories. Instead, our match and
mismatch names (such as car, goat, typewriter, penguin) reflected
the types of responses we observed in our own naming experiments
and in Snodgrass and Vanderwart (1980). However, our results
should still apply to basic-level object recognition, because the basic
and entry levels for common objects tend to overlap and because
many of our descriptions used basic-level names.
word UPSIDE
subjects, the description first specified orientation (e.g., UPSIDE
DOWN TREE). For the remaining subjects, the description first speci-
fied identity (e.g., TREE UPSIDE DOWN). The subject continued the
-DOWN, and the name of a familiar object. For half the
-
-
trial by pressing a mouse button, which erased the entire display and
triggered the object to appear 500 msec later. The subjects were in-
structed to respond as quickly as possible without making errors.
Response timing began with the onset of the picture; the duration of
the picture was equal to the response time.
A Latin-square design was used to fully counterbalance objects
across subjects with the 4 (match conditions) 3 2 (depicted orien-
tations) 3 2 (description formats) 5 16 conditions. Each subject
was assigned a unique trial schedule and, therefore, participated in
a single block of 96 trials comprising 96 different objects. For each
subject, (1) the order of trials within the schedule was randomized
without constraint, and (2) mismatch names were randomly selected
without replacement from a list of names of nondepicted objects.1
Results
The mean correct RTs and error rates from each group
of subjectswere submittedto a two-way, mixed analysisof
variance (ANOVA) to measure the within-subjects effect
mismatch
of
(orientation,identity, or global), the between-
The subscripts RT and ER are used throughout to denote analy-
ses based on RTs and error rates, respectively. The p values reported
description order
subjects effect of
(orientation–name or
for effects involving within-subjects factors are based on the name–orientation), and the interaction. (Excluding the
Geisser–Greenhouse lower bound adjustment to degrees of freedom
(e.g., Lewis, 1993).
yes
global match [ ] data from the ANOVA ensures that re-
sponse effects cannot confound the results. However, RTs
and error rates for all four match conditionsare presented
in Table 1 for completeness.)
EXPERIMENT 1
Replication and Further Analysis
Effects ofdescriptionorder. The resultsoftheANOVA
showed that description order had no effect on perfor-
We will begin with a replication of the original mance. When orientationwas specified first, the mean RT
description–picture experiment, in which subjects are and error rate were 569 msec and 3.3%, respectively, as
tested with upright (0º) and upside-down (180º) novel de- compared with 571 msec and 3.1% when name was spec-
F
,
MS 5
F
,
pictions of common objects. The procedures, items, and ified first [ RT(1,14) 1,
MS 5 p .
30,757.67, ER(1,14)
e
equipment were identical to those described in DeCaro 1,
4.70, s .75]. These null effects can be inter-
(1998),exceptthathere, weincludethepreliminarygrouped preted unambiguously because description order did not
MS 5
e
UPRIGHT CAR
CAR
F
,
data, in which descriptionorder (e.g.,
vs.
interact with type of mismatch [ RT(2,28) 1,
MS 5 p .
e
UPRIGHT
F
,
) was tested across two groups of 8 subjects. This 1,858.99, ER(2,28) 1,
6.72, s .40]. In light
e

E
814
D CARO AND REEVES
Table 1
cross model,since thevariabilitymust be sufficientto show
crossing.
The smooth curves in Figure 2 are logistic functions fit
to the mean data for the two partial-mismatch conditions.
We used curve fitting to extrapolate beyond the reaction
limits of our subjects, where any crossing is likely to be
most evident. This particular psychometric function was
Mean Correct Response Time (in Milliseconds), Standard
Deviation, and Percentage Error as a Function of Identity
and Orientation Match in Experiments 1–3A
Orientation
Match
Mismatch
Identity
M
SD
%E
M
SD
%E
a
chosen for its simplicity: Parameter controls the loca-
b
tion, and controls the slope (or shape). A higher value
Experiment 1 (n 5 16)
2
Match
Mismatch
580
517
133
84
9.9
1.0
675
518
125
96
7.8
0.8
b
of corresponds to a steeper cumulative function and,
Experiment 2 (n 5 12)
therefore, to decreased variabilityinthegeneratingprocess.
The ordinate in Figure 2 is the cumulative proportion of
subjects’ total correct judgments. This measure, rather
Match
Mismatch
536
535
99
116
6.9
3.1
667
513
118 10.8
84
0.3
proportion correct
than , was used so that each subject’s
Experiment 3A (n 5 16)
Match
Mismatch
616
532
141
91
7.2
0.6
705
516
149
73
7.9
0.6
two curves would always extend from 0 to 1, permitting
precise comparisonsof encodingrates and curve locations
despite differing error rates.
Consistent with the grouped data plotted in Figure 2,
the individualpsychometric functions fit to each subject’s
data showed that the rate of encoding for orientation was
slower than the rate of encoding for identity (in 12 out of
of these results, the orientation–name format was used in
subsequent experiments, because it conforms to the
adjective–noun grammar of the English language and
does not appear to bias the outcome.
t
16 cases). This was confirmed by a paired test of the 32
b 5
estimates of the slope parameters, whose means,
5.83
Effects of mismatch. Type of mismatch had large ef-
b 5
(for orientation) and
nificantly [ (15)
7.20 (for identity), differed sig-
F
5
MS 5
fects on both RTs and errors [ RT(2,28) 70.80,
e
t
5
p ,
2.62, .05]. However, the times
F
5
MS 5
p ,
6.72, s .001].
1,858.99,and ER(2,28) 37.85,
The planned comparisons between identity mismatches
e
needed to verify orientation did not vary enough to main-
tain the supposition that orientation is determined before
identity. As is shown in Figure 2, any difference in slope
M 5
M 5
(
(
517 msec, 1.0% errors) and orientationmismatches
675 msec, 7.8% errors) showed that the subjects
were faster and more likely to correctly verify a mismatch
F
5
in identity than a mismatch in orientation [ RT(1,28)
MS 5 MS 5
F 5
106.90,
1,858.99, and ER(1,30) 54.60,
e
e
p ,
6.72, s .001]. Moreover, it is evident from Table 1 that
responses to identity mismatches were as fast and as ac-
F ,
curate as responses to global mismatches ( s 1).
Psychometric Analysis and Discussion
The results of Experiment 1 were consistent with those
reported in DeCaro (1998) in both the magnitudesand the
patterns of effects on RTs and error rates. Table 1 shows
different-identity
that performance on
trials (i.e., identity
mismatch and global mismatch) did not vary with orien-
tation match. Thus, the increased RTs on orientation mis-
match trials can be attributedto the time needed to extract
a viewer-centered descriptionof the object—for example,
to locate “top” (Rock, 1973). If so, orientation encoding
entirelyfollowsextractionof objectclass (the shift model).
It is, however, still possible that the object’s orientation is
extracted before the object has been identified but that
identity is encoded more rapidly and, so, will eventually
give rise to faster object identification. Such early orien-
tation encoding may even work to propel the process of
objectidentificationby guidingtransformationsof rotated
objects (e.g., Tarr & Pinker, 1991). The cross model im-
plies that the underlying distributionof latencies to verify
orientation is more variable than is the distribution of la-
tencies to verify identity. However, such increased vari-
Figure 2. Mean proportion of total mismatches detected as a
function of time in Experiment 1. Psychometric curves are best-
fitting logistic functions, y 5 (x / a) / [1 1 (x / a) ], where a is the
response time corresponding to 50% correct and b determines
the slope. The response times above 1,000 msec were too few and
too scattered to include in this plot. Error bars are 61 standard
b
b
ability in the observed RTs would not by itself prove the error.

OBJECT IDENTIFICATION AND ORIENTATION
815
was countered by a large difference in the locationsof the served as descriptions on the mismatch trials. Hence, none of the 96
name probes were presented to the subject prior to the recognition
test, but half of the names referred to objects that were depicted in the
description–picture task. After entering each yes/no response, the
subjects rated how confident they were in their decisions, using the
following scale: 0 5 don’t know (guessing); 1 5 barely confident;
2 5 reasonably confident; 3 5 very confident.
a
two functions: The respective mean values of were 616
p ,
t
5
8.94,
and 494 msec [ (15)
.001]. That the two
curves in Figure 2 do not cross shows that the increase in
RTs for verifying orientation probably reflects a genuine
shift
, in the time to encodeorientation—not merely
lag, or
a slower rate of processing.
Results and Discussion
EXPERIMENT 2
Recognition for Mismatch Objects
The results of the description–picture task are summa-
rized in Table 1. Mean correct RTs and error rates were
submitted to a one-way repeated measures ANOVA to
measure the effects of mismatch, which were significant
It is assumed that name verification reflects a fairly ef-
ficient process of object recognition. However, consider
the identity mismatch trial illustrated in Figure 1. Hamm
and McMullen (1998) point out that, in such a case, the
depiction is identified minimally as “not a . . .” (in this
case, “not a rabbit”). It is known that objects that share a
basic level tend to be visually similar, whereas objects
from different basic categories are often clearly distin-
guishable in terms of global shape (Rosch et al., 1976).
Given the heterogeneity of the objects in our study, it is
likely that global shape information would have been suf-
ficient to signal an identity mismatch on many trials,
thereby obviating the need for classification at the more
specific entry level. Indeed, the subjects may have suc-
cessfully rejected mismatch objects by comparing the de-
pictions and mismatch names at the superordinate level
F
5
MS 5
p ,
s
F
5
[
RT(2,22) 43.81,
MS 5
1,910.17, and ER(2,22)
e
33.75,
M 5
10.35,
535msec, 3.1% errors) were verified faster and more
M 5
.001]. Identity mismatches
e
(
accurately than orientation mismatches [
10.8% errors; RT(1,22) 54.87,
667 msec,
1,910.17, and
10.35, s .001], but some-
M 5
F
5
MS 5
e
F
5
MS 5
p ,
ER(1,22) 33.84,
e
what less accurately than global mismatches [
0.3%
p 5
10.35, .06]. This
F
5
MS 5
errors; ER(1,22) 4.48,
marginal global mismatch effeect on error rates resulted
from a slight increase in the number of errors on identity
mismatch trials, as compared with the number of errors
we obtained in Experiment 1. However, the results again
F
5
showed no global mismatch effect on RTs [ RT(1,22)
MS 5 p 5
1.57,
1,910.17,
.24].
e
Individual scores on the recognition test ranged from a
low of 61.5% correct to a high of 90.6% correct, with a
mean of 74.6%, which was well above chance perfor-
animal
(e.g., “the depicted object is not an
not a rabbit”).
and, therefore,
Our first approach to testing the specificity in
description–picture matching was to measure the subject’s
recognition of mismatch objects. To illustrate this ap-
t
5
p ,
mance [ (11) 9.14,
.001]. The lowest score, 61.5%,
z
p ,
is itselfsignificantlybetter than chance (
5
2.35,
.05,
by normal approximationto the binomial).Thus, guessing
can be rejected as the sole explanationof performance for
all 12 subjects.
tree
proach, suppose that a subject viewed the picture of a
without the word “tree” in the matching task, as in Fig-
ure 1. Further suppose that the subject correctly reported
a mismatch, because the picture did not match the pre-
ceding “rabbit” description (i.e., not a guess). The ques-
tion is whether the picture was identified merely as “not a
rabbit.” If yes, it would be impossibleto recall havingseen
a tree when probedwitha word stimulus,“tree” (rather than
the original depiction), because our pictures and words
can match only at the entry or basic levels. Data inconsis-
tent with guessing therefore provide partial support that
mismatch objects are verified at a relatively high level of
specificity.
M 5
The confidenceratings for correct responses (
2.46)
were significantly higher than the ratings for incorrect re-
M 5 p ,
.001], implyingthat
t
5
sponses [
2.03; (11) 4.84,
the subjects were somewhat aware of their own errors but
were adopting a conservative response criterion. To sepa-
rate memory performance and response bias, each sub-
yes
ject’s hit rate (
to a probe naming a depicted object)
yes
d
¢
and false alarm rate (
to a foil) were converted to
. The provides a measure of stimulus discrimina-
tion that is independentof response bias, (Macmillan &
Creelman, 1991). Values of greater than 1 indicate a
no
d
¢
and
b
b
b
bias toward responses, and indeed,the average value of
in our experiment was 3.02. This agrees with the self-
Method
b
Subjects. Twelve undergraduate students participated in Experi-
ment 2.
rating data and confirms that the subjects tended to as-
sume that they had not seen any of the objects named in
the recognitiontest unless the name probe eliciteda strong
sense of familiarity.
Recognition sensitivity (as measured by ) ranged
from a low of 0.87 to a highof 2.92, with a mean over sub-
Procedure. The equipment, materials, and procedures were iden-
tical to those used in Experiment 1, but the subjects were also given
a surprise recognition test about 2 min after completing the
description–picture matching trials. During the recognition test, the
subjects were probed with the names of common objects and were
asked to press the “Y” key if a name referred to an object depicted
in the description–picture task, and to press the “N” key otherwise.
The probes consisted of the 48 names of objects that had been de-
picted on identity mismatch and global mismatch trials and 48
d
¢
t
5
jects of 1.60, well above the chance level of zero [ (11)
p ,
8.58,
.001]. When hit and false alarm rates were first
pooled d
¢
averaged across subjects instead, the resulting
foils—items from the list of mismatch names that had not already was 1.44, still well above zero. We should emphasize that

E
D CARO AND REEVES
816
up down right
, and
left
to ob-
the subjects completed the description–picture matching sociate the spatial words
,
,
trials not knowing that they would have to remember the jects presented at 0º, 180º, 90º, and 270º, respectively. In
depicted objects (all of which were intermixed with po- a subsequent word–picture verification task, the subjects
tentially interfering object names). These results indicate saw a spatial word followed by a picture of an object. The
intactrecognitionfor entry-levelidentity(at least for most results showed that matches were reported faster than mis-
subjects) and rule out the possibility that mismatch ob- matches when pictures were upright, whereas matches
jects were identified as “not a . . .” and nothing more spe- were reported slower than mismatches when pictures were
cific. Thus, the findings of Experiment 2 support the con- upside-down. Maki and Braine attributed this interaction
clusion of Experiment 1 (DeCaro, 1998) that identity is effect to response compatibility, resulting from an implicit
true
encoded before orientation.
tendency to associate upright with
and upside-down
false
In word–picture matching tasks, it may be possible to with
. If so, direct comparisons between upright and
reject a mismatch object on the basis of prominent visual rotated depictions in the description–picture procedure
features, such as global shape. Yet the findings of Exper- may be confounded by one or more nuisance variables.
iment 2 favor the conclusion that the subjects rejected Our solution was to add multiple rotated views to the de-
mismatch objects at a semantic level—specifically, the sign without changingthe task. By comparing the partial-
within
entry level. It is supposed that the subjects retrieved the mismatch trials
the rotated-view conditions, it is
entry-level identity of the depicted object and then com- then possible to properly measure the effects of angular
pared it with the description, also at the entry level. Dur- rotation on verification times.
ing the surprise recognition test, a positive name probe
would activate the same entry-level concepttriggeredear- Method
Subjects. Sixteen undergraduate students participated in Exper-
iment 3A.
Stimuli and Equipment. The set of pictures was increased to 128
lier by the picture, signalingthat the name referred to one
of the depicted objects.
line drawings of common objects, which included the items used in
EXPERIMENT 3A
Verifying Upright and Multirotated Objects
Experiments 1 and 2. Misoriented versions of each object were cre-
ated by rotating the standard version 45º, 90º, 135º, and 180º in the
picture plane. The equipment was as described in Experiment 1.
Procedure. The procedures were generally the same as those in
Experiment 1, with a few exceptions. First, because misoriented ob-
The verification times in the description–picture task
were consistently compatible with earlier encoding for
identity, rather than for general orientation, which pro-
vides partial support for view-invariantaccounts of visual
object recognition.Using a backward-maskingprocedure
jects were not always rotated 180º, the description UPSIDE-DOWN was
changed to ROTATED. Second, for each subject, 32 unique objects
were randomly assigned to each of the four match conditions. Within
each match condition, 50% of the items were randomly selected to
appear at 0º, and the remaining objects were randomly selected to
appear at one of the four rotated views, with the constraint that an
equal number of objects was assigned to each view. The subjects
were instructed that any nonupright object should be classified as ro-
tated when matching orientation— since the degree of rotation did
not matter. The final ordering of trials was completely random,
and—as in the first two experiments—the objects were presented
without repetition.
picture word
verification,DeCaro and Reeves (2000)
and
–
also showed that identity was determined before orienta-
tion and, in addition, that processing time to extract iden-
tity was relatively flat from 60º to 180º of rotation. Hamm
and McMullen’s (1998) word–picture experiment varied
collie
dog
col-
the level of classification of a match (e.g.,
vs.
) and showed that subordinate-level (e.g.,
animal
vs.
lie
) decision times were most affected by rotations from
the upright, whereas basic-level and superordinate-level Results and Discussion
decision times were about equally least affected—
although still not flat through 180º of rotation.
The mean correct RTs and error rates were submittedto
a two-way repeated measures ANOVA to measure the ef-
mismatch
view
and the interaction. The main
Whereas the word–picture procedure used by Hamm fects of
and
F
5
and McMullen (1998) probed the classification of object effects of mismatch were significant [ RT(2,30) 45.16,
MS 5 MS 5 p ,
F
5
class at different levels of specificity and varied orienta-
13,503.15, ER(2,30) 15.11,
76.37, s
e
e
tion in the picture plane, the description–picture proce- .001].Table1 shows that these effects were consistentwith
dure used by DeCaro (1998) probed the classification of thoseobtainedin Experiment1, despitethechangeto multi-
both object class and orientation,but it could not properly rotated views. The main effects of view were not signifi-
F
5
MS 5
p .
F
10,589.36,and ER(4,60)
5
´
measure the effects of rotations in the plane. First, there cant [ RT(4,60) 0.87,
are only two spatial views, but more views are required to 1.36,
e
MS 5
67.73, s .25]. However, the mismatch
test for progressive effects of rotation. Second, the RTs to view inteeraction was marginally significant even with the
F
5
verify orientationmismatcheswhen objectsare uprightre- conservativeadjustmentto degreesof freedom [ RT(8,120)
not rotated
MS 5
p 5
flect the time to determine that objects are
.
3.26,
7,239.72, .09; there were too few errors
e
Conversely, when objects are rotated, the RTs reflect the at this level of analysis to test for effects on error rates].
not upright
We therefore reexamined the effects of view on just the
time to determine that objects are
An experiment by Maki and Braine (1985) suggested partial-mismatch trials, using post-hoc contrasts.
that there may be encoding differences between these two
Figure 3 shows that the difference in RTs across the ori-
types of classifications. Their subjects first learned to as- entation mismatch and identity mismatch trials varied
.

OBJECT IDENTIFICATION AND ORIENTATION
817
We therefore reasoned that the difference between up-
right and nonupright views on identity mismatch trials
shown in Figure 3 is an artifact of spatial word, because
upright
rotated
for
the descriptionwas
for 0º views and
the 45º–180ºviews (owing to the orientationmatch). If so,
the global-mismatchcases shouldhave manifestedthe op-
posite effect (owing to the orientation mismatch). When
the global mismatch cases were also classified according
to the object’s physical orientation, such a pattern was
found. Consequently, the identity mismatch and global
mismatch cases were submitted to a new ANOVA, but the
word
data were explicitly classified according to spatial
upright
rotated)
view
(
vs.
and depicted (0º to 180º), ig-
noring whether there was a match or a mismatch in orien-
tation. The ANOVA confirmed a significant main effect
F
5
MS 5
p ,
3,755.38,
of spatial word [ RT(1,15) 23.36,
.001]. As was predicted, RTs were fasteer on trials with the
upright M 5
word
(
505 msec) than on trials with the word
rotated M
(
5
542 msec). Neither the main effect of view
F
5
F
MS 5
3
[
RT(4,60) 1.68,
4,572.20]nor the word view
5 MS 5
e
interaction [ RT(4,60) 0.89,
p .
4,083.51] was sig-
e
nificant ( s .20).
Figure 3. Mean times to correctly verify identity and orienta-
tion mismatches as a function of spatial orientation (Experi-
ment 3A). Error bars are 1 standard error.
The veridicaleffect of rotation on times to verify object
identityis shown by combiningthe identitymismatch and
global mismatch cases, thereby canceling the effect of
spatialword. Corrected matchingtimes across the 0º–180º
range of views were 511, 534, 534, 553, and 524 msec.
This effect of plane rotation on RT was minimal, amount-
ing to 0.1 msec/deg. If orientation encoding or mental ro-
tationhad precededidentification,onemight haveexpected
a robust orientation effect with our novel (nonrepeated)
depictions (Jolicœur, 1990; Jolicœur & Milliken, 1989).
Experiment 3A produced two important findings. The
first was that rotationsin the picture plane did not increase
with angular view. First, on identity mismatch trials, 0º
views were verified faster than the four rotated views
(45º–180º), whereas on orientation mismatch trials, 0º
views were verified slower than the four rotated views
F
5
MS 5
p ,
.001]. Second,
[
RT(1,120) 21.57,
7,239.72,
e
times to verify identitywere essentiallyflat across the four
F
5
MS 5
p 5
with
rotated views [ RT(1,120) 0.08,
.78], whereas times to verify orientation
7,239.72,
e
decreased
F
5
increasing rotation from the upright [ RT(1,120) 4.96,
MS 5
p ,
7,239.72,
.05], as was shown in tests of the lin-
earetrend acrossthe45º, 90º, 135º, and 180ºviewconditions.
A separate analysiswas carried outto determinewhether
upright/rotated interacted with mismatch/match. Our re-
sults replicated Maki and Braine (1985), in that global
matches were verified faster than mismatches when ob-
jects were upright, whereas global matches were verified
slower than mismatches when objects were rotated. How-
ever, the response compatibility hypothesis (associating
true
false
and upside-down with ) cannot ac-
upright with
count for the interaction. The left side of Figure 4 shows
the match and mismatch cases plotted as a function of
whether objects were upright or rotated in the display.
Response compatibilitypredicts that mismatch responses
to a rotated object will be faster than mismatch responses
to an upright object. The critical result in Figure 4 (left) is
D
that the oppositeeffect occurred on identitymismatch ( )
trials. When the data were, instead, classified according to
spatial word, as shown on the right side of Figure 4, the
rotated
cause was obvious: Times to verify
than times to verify
were longer
upright
, regardless of match and re-
gardless of view. An inspection of the data from Experi-
upside
Figure 4. The effects of spatial orientation (left panel) and spa-
ment 1 showed the same effect with the words
down upright
-
ord (right panel) on times to correctly verify matches and
tial w
and
.
mismatches in Experiment 3A.

E
D CARO AND REEVES
818
the time required to determineobject identity. On the con- fied in Block 1 (Murray, 1995), and orientationpriming is
trary, the time needed to determine orientation decreased not expected when repeated objects are exclusively up-
with rotationup to 180º, presumably because it is easier to right in Block 1 (Jolicœur & Milliken,1989). The assump-
rotated
classify a misoriented object as
the further it is tion that objects are verified at the entry level in the
transformed from the canonicalview. However, more time description–picture task and the fact that objects are to be
was still required to determinethat an upside-down (180º) depictedat severalorientationstogetherpredictbothobject-
object was rotated than to determine the object’s identity. priming and orientation-primingeffects on naming times
Thus, the conclusion from Experiment 1 that identitypre- in Block 1.
cedes orientation is true regardless of whether subjects
must detect small (45º), medium (90º), or large (135º and Method
Subjects. The subjects were 32 undergraduate students, 16 of
whom were the same as those in Experiment 3A. The other 16 sub-
jects were controls, having not participated in any of the previous ex-
periments.
180º) rotations in the picture plane.
This experiment also demonstrated the influence of
spatial words in word–picture matching tasks. The results
showed that the interactionbetween spatialdimension and
Design. The control subjects completed two blocks of naming tri-
match was due to a main effect of spatial word, and not to
response compatibility. The details in Maki and Braine
(1985) showed that their interaction effect was also con-
sistent with a simpler main effect of word. We therefore
choose the more parsimonious spatial word effect to ex-
als (denoted B1-control, and B2-control, respectively), whereas the
test subjects from Experiment 3A completed only a single block of
naming trials (B1-test). We expected transfer from Block 1 naming
to Block 2 naming with the control subjects, owing to the two doc-
umented priming effects. The purpose of this design was to discover
whether priming would also transfer from the matching task to nam-
ing with the test subjects. If so, naming times in the B1-test condi-
tion should more closely follow the naming times in the B2-control
condition.
upright
plain both results. The fact that
upside-down rotated
can be verified
points to differences
faster than
and
in the encodingand interpretationof certain spatiallabels.
These differences should be consideredwhen interpreting
performance in word–picture tasks that involve spatial
judgments, such as location, position, and orientation.
The items used on B1-control trials were identical to those used
on the description–picture trials in Experiment 3A. Each control
subject was paired with one data file from Experiment 3A. These
files contained records of the objects and views that were used on
each description–picture trial. A computer program read this infor-
mation and then generated equivalent trial schedules for the control
subjects. In addition, the items used on B2-control trials were iden-
tical to those used on B1-test trials. These two critical blocks com-
prised only the previously rotated objects (for the test subjects, these
were all the rotated objects presented in the description–picture
task). It was necessary to exclude the 50% of the objects previewed
at 0º because repeated objects generally do not produce orientation
priming in Block 2 if they were upright in Block 1 (Jolicœur & Mil-
liken, 1989). Each of the 64 previously rotated objects was randomly
assigned to appear at a new orientation (including 0º), with the con-
straint that an equal number of objects was assigned to the five
views. A different randomization sequence was used for each test
subject and then duplicated for the corresponding control subject.
Procedure. The equipment used to present objects on the nam-
ing trials was the same as that described in Experiment 1. A black
ring appeared at the center of the display to signal the start of each
trial. The subject then pressed a mouse button, which blanked the
display and caused the picture to appear (until named) 500 msec
later. The subjects were instructed to name each object as quickly
and clearly as possible. A microphone, voice-activate d relay, and soft-
ware timer measured naming latencies to the nearest millisecond.
The sensitivity of the voice-activated relay was calibrated to each
subject during an initial practice block, while the subject named a se-
ries of alphanumeric characters. The test subjects were run approx-
imately 5 min after completing the description–picture experiment
(the time needed to complete the practice trials and voice calibra-
tion).
EXPERIMENT 3B
Object Naming
The addition of multirotated objects in Experiment 3A
provided an opportunity to test the specificity of our
description–picture task in a completely different manner
than in Experiment 2. In this approach, we sought to un-
cover processes common to word–picture matching and
transfer
—that is, for
objectnaming by testing for intertask
changes in naming performance resulting from word–
picture verification.The followingpointswill help to clar-
ify and motivate this use of transfer. When rotated objects
are repeated in a naming experiment, there are two promi-
nent markers that distinguish performance on Block 2
from that on Block 1 (see Jolicœur, 1985; Maki, 1986).
object priming
: Repeated objects are named
First, there is
faster in Block 2 than in Block 1. Second, there is
orien-
tation priming
: The size of the orientationeffect is smaller
in Block 2 than in Block 1, yielding a much flatter nam-
ing function. These two markers show priming, rather
than mere practice, because they disappear when new ob-
jects are introduced in later blocks of trials (Jolicœur,
1985).
In principle,transfer of these two priming effects can be
used to test convergence between word–picture matching
and object naming. Objects are typically named with Results and Discussion
entry-level terms (Snodgrass & Vanderwart, 1980), the
Naming times longer than 3 sec were discarded as obvi-
same level as that used in our descriptions. If objects are ousoutliers,whichaffected 0.7%ofthedata.Incorrectnam-
verified
at the entry level, priming should transfer from ing responsesand spoiled trialswere also discarded, which
matching to naming, manifesting itself in the very first, excluded 8.0% of the data. A trial was considered spoiled
rather than the second,blockof namingtrials. Object prim- if the naming response failed to activate the voice key or
ing is not expected when repeated objects are not identi- if a vocal hesitation activated the voice key prematurely.

OBJECT IDENTIFICATION AND ORIENTATION
819
Table 2
Mean Correct Naming Time (in Milliseconds) and Standard Deviation
for Each Type of Block in Experiment 3B
B1 Test
B1 Control
SD
1,070 130
B2 Control
I1/O1
I+/O2
I2/O1
I2/O2
M
M
SD
M
SD
M
SD
M
SD
M
SD
888
111
874
144
874
141
890
140
922
187
Note—The data from test subjects who completed the description–picture task priorto naming
objects are broken down to show in which match condition the objects were first presented: I
and O denote identity and orientation, respectively, and + and 2 denote match and mismatch,
respectively.
Control priming. The data from the control subjects jects. The findings of Experiment 3B thus strengthen the
were submitted to a two-way repeated measures ANOVA interpretation of results from the surprise recognition test
to measure the effects of block and view and the interac- in Experiment 2. Taken together, a more convincingargu-
F
5
tion. The main effect of block was significant [ (1,15)
ment can be made that match and mismatch objects were
MS 5
p ,
8,833.79,
150.18,
faster in Block 2 (
.001]. Naming times were verified at a semantic level consistent with entry-level
e
M 5
M 5
888 msec) than in Block 1 (
classification.
GENERAL DISCUSSION
1,070 msec). The main effect of view was also significant
MS 5 p ,
F
5
[ (4,60) 10.75,
5,696.24,
.005]. Naming
e
times increased with rotation from the upright, but with a
departure in linearity at 180º that is typical in rotated ob-
In threeexperiments,we haveattemptedagain todemon-
strate that object class is determined even before general
orientation (upright vs. misoriented) is known (DeCaro,
3
5
ject naming (e.g., Jolicœur, 1985). The overall block
F
view interaction was not reliable in this case [ (4,60)
MS 5 p 5
2.37,
9,442.23,
.14]. However, the slope of the 1998) and that the recognition of plane-rotated views of
e
orientation effect was smaller in Block 2 than in Block 1, common objects does not require a prior stage of mental
as was shown by a significant linear interaction compo- rotation (Jolicœur, Corballis, & Lawson, 1998) or an in-
F
5
MS 5
p ,
9,442.23, .05]. Thus, cremental normalizationprocess that mimics mental rota-
nent [ (1,60) 7.31,
these data are consistent weith previous object-namingex- tion. The psychometric functions in Experiment 1 show a
periments showing both object priming and orientation clear shift and indicatethat, using RT this time, orientation
priming in Block 2.
is not merely encoded later than identity at all levels of
Test priming. Table 2 shows the mean correct naming performance: It accrues more slowly. We have suggested
times for each type of block. The B1-test trials are broken that a mental rotation-likeprocess is used to determine(or
down to show the specific match condition in which the
objects were previewed (in Experiment 3A) prior to nam-
ing. It is clear from these data that naming times were
M 5
shorteronB1-test trials(
890msec) than onB1-control
M 5
t
5
p ,
trials [
alsoshowsthatthefourmatch conditionsinthedescription–
MS 5
1,070 msec; (30) 3.76,
.001]. Table 2
F
5
picture task primed equally well [ (3,45) 1.53,
p 5
e
5,450.59,
.23].
Next, data from the B1-test and B2-control conditions
were submitted to a mixed two-way ANOVA with block
and view as the between- and within-subjects factors, re-
spectively. The two groups of subjects showed equivalent
F
5
priming: There was an effect of view [ (4,120) 5.10,
MS 5 p ,
7,370.62,
.05], but neither the effect of block
MS 5
e
F
5
3
[ (1,30) 0.03,
82,745.61]nor the block view
5 MS 5
e
F
interaction [ (4,120) 0.64,
p .
7,370.62] was sig-
e
nificant ( s .40). Trend analysis showed that any linear
p 5
interaction component also was unreliable (
.19).
The substantive importance of these results is obvious
from Figure 5: Verifying objectsin the description–picture
task produced significant priming effects in a subsequent
object-naming task. We obtained both object priming and
orientation priming effects in Block 1 that were compara-
Figure 5. Mean correct naming times as a function of spatial
ble to the effects normally seen in Block 2 with other sub- orientation for each block in Experiment 3B.

E
820
D CARO AND REEVES
check
picture–word
matching(DeCaro & Reeves,
) orientationin the picture plane (DeCaro & Reeves, procedure and
2000), but our data in Figure 3 imply a rotation speed of 2000). Thus, we have argued, like Corballis (1988), that
1,538 deg/sec (or 0.65 msec/deg), much too fast for men- the effects of orientation on naming RTs occur at a
postrecognition
tal rotation. The very rapid transformations implied by
stage of double-checking. Familiarity
these data might reflect the fact that our task required ori- with the misoriented material is sufficient to virtually
entation judgments at only the lowest level of specificity: eliminate orientation effects on naming RTs for common
Simply determiningthat the picture is not uprightwas suf- objects—unlike the effects on times to judge orientation,
ficient classification. If we had required subjects to clas- which are linear and robust across practice (e.g., Cooper
sify specific views (such as 45º vs. 90º vs. 135º), we might & Shepard, 1973; Corballis & Cullen, 1986; Jolicœur, In-
have observed much slower rates, consistent with earlier gleton, Bartram, & Booth, 1993; McMullen & Jolicœur,
mental rotation studies (e.g., Cooper & Shepard, 1973). 1990, 1992). We therefore argued that the increase in RTs
The orientation task was designed to be overly simple (to associated with rotations in the picture plane can be at-
bolsterthe identity-before-orientationcase), and so the ro- tributed quite reasonably to the use of view-dependent
orientation
of
tation curves may reflect this ease.
Perhaps more important, in light of the results already newly rotated objects (DeCaro & Reeves, 2000).
established, we showed how an unconventional word–
Our claims regarding view dependencyin visual object
mechanisms to check or determine the
picture matching task could be used to probe visualobject recognition apply to line-drawing depictions of common
recognition. DeCaro’s (1998) method implies a high pro- objects. That such drawings can support recognition al-
portionof mismatch trials (here, 75%) so that latenciesfor most as well as full-color representations (Biederman &
no
the same response ( ) can be used as a probe across dif- Ju, 1988) implies a powerful perceptual engine that can
ferent conditions. This has the enormous advantage that abstract critical features for recognition from line seg-
incidental factors that can influence responses in same– ments, intersections,junctions,and so forth. Such critical
different paradigms (such as handedness,frequency of re- features are heavily dependent on correct analysis of rel-
sponse, and number of matching elements) cannot affect ative orientation (e.g., that one line segment is clockwise
the conclusions and, so, need not be controlled (see 15º relative to another, or that two lines meet at a right
Krueger, 1973;Nickerson, 1967).Still, a possiblecriticism angle). A rotation in the picture plane, of course, changes
mismatches
is that
can be determined on open-ended,neg- none of these geometric relations, but such transforma-
X
perceptual effect
ative grounds (“not an ”), and if so, they would not ad- tions can have a strong
on global shape
dress object recognition at the entry level. A critical find- (Rock, 1973, 1974) and may affect particular features of
ing, however, is that most of the subjects demonstrated the image. Our present conclusionis that double-checking
intact memory for mismatch object names in a surprise (for orientation) is still a valid hypothesis to explain how
recognition test following the description–picture trials, early, orientation-independentrecognitionmight deal with
showing that the subjects had classified the objects in a the perceptual consequencesof common objects depicted
positive manner. In early pilot work, we obtained similar at uncommon views.
recognitionscores when picturedurationinthe description–
picture task was limited to 55 msec with backward mask-
REFERENCES
ing, which makes it less likely that memory performance
Biederman, I., & Ju, G. (1988). Surface versus edge-based determi-
reflectsanyencodingbeyondthatneededtomake a speeded
nants of visual recognition. Cognitive Psychology, 20, 38-64.
response in the description–picture task.
Anotherpossiblecriticism is that the task set to subjects
in word–picture verification, a simple
Cooper, L. A., & Shepard, R. N. (1973). The time required to prepare
for a rotated stimulus. Memory & Cognition, 1, 246-250.
Corballis, M. C. (1988). Recognition of disoriented shapes. Psycho-
logical Review, 95, 115-123.
same–different
judgment, is so different from naming that word–picture
matching performance cannot bear on naming RTs. Thus,
the finding by Hamm and McMullen (1998), that basic-
level judgments showed weak effects of orientation, and
our second critical finding, that times to verify identity
were flat across 180º of rotation, may not contradict the
classic naming experiments of Jolicœur (1985) and Maki
(1986), which do show a clear, systematic effect of orien-
tation on times to name similar objects. However, in Ex-
periment 3B, we showed direct transfer of priming from
word–picture matching to naming, demonstrating an im-
portant convergence in the encoding and recognition
processes brought to bear on the two tasks.
Corballis, M. C., & Cullen, S. (1986). Decisions about the axes of
disoriented shapes. Memory & Cognition, 14, 27-38.
Corballis, M. C., Zbrodoff, N. J., Shetzer, L. I., & Butler, P. B.
(1978). Decisions about identity and orientation of rotated letters and
digits. Memory & Cognition, 6, 98-107.
DeCaro, S. A. (1998). On the perception of objects and their orienta-
tions. Spatial Vision, 11, 385-400.
DeCaro, S. A., & Reeves, A. (2000).Rotating objects to determine ori-
entation, not identity: Evidence from a backward-masking/dual-task
procedure. Perception & Psychophysics, 62, 1356-1366.
Graves, R., & Bradley, R. (1987). Millisecond interval timer and au-
ditory reaction time programs for the IBM PC. Behavior Research
Methods, Instruments, & Computers, 19, 30-35.
Graves, R., & Bradley,R. (1991). Millisecond timing on the IBM PC/
XT/AT and PS/2: A review of the options and corrections for the
Graves and Bradley algorithm. Behavior Research Methods, Instru-
ments, & Computers, 23, 377-379.
An argument can always be made that naming is spe-
cial, but we have also found direct evidence that rotation
in the plane has systematic effects on the encoding of ori-
entation, rather than identity, using a backward-masking
Hamm,J. P.,& McMullen,P. A. (1998).Effects of orientationontheiden-
tificationofrotated objectsdependon thelevel ofidentity. Journalof Ex-
periment Psychology: Human Perception & Performance, 24, 413-426.

OBJECT IDENTIFICATION AND ORIENTATION
821
Jolicœur, P. (1985).The time to name disorientednatural objects. Mem-
ory & Cognition, 13, 289-303.
Jolicœur, P. (1990). Identification of disoriented objects: A dual-
systems theory. Mind & Language, 5, 387-410.
Nickerson, R. S. (1967). “Same”–“different” response times with
multi-attributestimulusdifferences. Perceptual & MotorSkills, 24, 543-
554.
Rock, I. (1973). Orientation and form. New York: Academic Press.
Jolicœur, P., Corballis, M. C., & Lawson, R. (1998). The influence Rock, I. (1974, January). The perception of disoriented figures. Scien-
of perceived rotary motion on the recognition of rotated objects. Psy-
chonomic Bulletin & Review, 5, 140-146.
Jolicœur, P., Gluck, M. A., & Kosslyn, S. M. (1984). Pictures and
tific American, 230, 78-85.
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-
Braem, P. (1976). Basic objects in natural categories. Cognitive Psy-
chology, 8, 382-439.
names: Making the connection. Cognitive Psychology, 16, 243-275.
Jolicœur, P., Ingleton, M., Bartram, L., & Booth, K. S. (1993). Segalowitz, S. J., & Graves, R. E. (1990). Suitability of the IBM XT,
Top–bottomand front–behind decisions on rotated objects. Canadian
Journal of Experimental Psychology, 47, 657-677.
AT, and PS/2 keyboard, mouse, and game port as response devices in
reaction time paradigms. Behavior Research Methods, Instruments,
& Computers, 22, 283-289.
Snodgrass, J. G., & Vanderwart, M. (1980). A standardized set of
260 pictures: Norms for name agreement, image agreement, famil-
iarity, and visual complexity. Journal of Experimental Psychology:
Human Learning & Memory, 6, 174-215.
Jolicœur, P., & Milliken, B. (1989). Identification of disoriented ob-
jects: Effects of context of prior presentation. Journal of Experimen-
tal Psychology: Learning, Memory, & Cognition, 15, 200-210.
Krueger, L. E. (1973). Effect of stimulus frequency on speed of
“same”–“different” judgments. In S. Kornblum (Ed.), Attention and
performance IV (pp. 497-506). New York: Academic Press.
Lewis, C. (1993). Analyzing means from repeated measures data. In
G. Keren & C. Lewis (Eds.), A handbookfor data analysis in the be-
havioral sciences: Statistical issues (pp. 73-94). Hillsdale, NJ: Erl-
baum.
Tarr, M. J., & Pinker, S. (1991).Orientation-dependentmechanisms in
shape recognition:Further issues. Psychological Science, 2, 207-209.
NOTES
Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: A
user’s guide. New York: Cambridge University Press.
1. There were minor variations in the lists of match and mismatch
names used throughout this study (for instance, the lists were expanded
Maki, R. H. (1986). Naming and locating the tops of rotated pictures. in Experiment 3), but most of the items are described in DeCaro (1998).
Canadian Journal of Psychology, 40, 368-387. Alternatively, the reader can e-mail the first author for a complete list of
Maki, R. H., & Braine,L. G. (1985).The role ofverbal labels in thejudg- the match and mismatch names used in the present study.
ment of orientation and location. Perception, 14, 67-80.
McMullen,P. A., & Jolicœur, P. (1990).Thespatial frame of reference
in object naming and discrimination of left–rightreflections. Memory
& Cognition, 18, 99-115.
McMullen,P. A., & Jolicœur, P. (1992).Reference frame and effects of
orientation on finding the tops of rotated objects. Journal of Experi-
2. This curve is a simplified form of a more general three-parameter
equation, in which x 5 (RT 2 c)/a permits scaling and shifting of the
RTs with two parameters, and then the logistic gives cumulative propor-
tioncorrect 5 xb / (1 1 xb), where b definesthecurveshape. Ourdata were
satisfactorily fit with c 5 0 (no shift), perhaps because our RTs were suf-
ficiently long. Had they been shorter, removal of the residual RT
mental Psychology: Human Perception & Performance, 18, 807-820. (namely, c) might have been necessary.
Murray, J. E. (1995). The role of attention in the shift from orientation-
dependent to orientation-invariant identification of disoriented ob-
jects. Memory & Cognition, 23, 49-58.
(Manuscript received November 22, 2001;
revision accepted for publication February 25, 2002.)