Automaticity of basic-level categorization accounts for labeling effects in visual recognition memory.
ABSTRACT Are there consequences of calling objects by their names? Lupyan (2008) suggested that overtly labeling objects impairs subsequent recognition memory because labeling shifts stored memory representations of objects toward the category prototype (representational shift hypothesis). In Experiment 1, we show that processing objects at the basic category level versus exemplar level in the absence of any overt labeling produces the same qualitative pattern of results. Experiment 2 demonstrates that labeling does not always disrupt memory as predicted by the representational shift hypothesis: Differences in memory following labeling versus preference are more likely an effect of judging preference, not an effect of overt labeling. Labeling does not influence memory by shifting memory representations toward the category prototype. Rather, labeling objects at the basic level produces memory representations that are simply less robust than those produced by other kinds of study tasks.
Automaticity of Basic-Level Categorization Accounts for Labeling Effects
in Visual Recognition Memory
Jennifer J. Richler, Isabel Gauthier, and Thomas J. Palmeri
Are there consequences of calling objects by their names? Lupyan (2008) suggested that overtly labeling
objects impairs subsequent recognition memory because labeling shifts stored memory representations of
objects toward the category prototype (representational shift hypothesis). In Experiment 1, we show that
processing objects at the basic category level versus exemplar level in the absence of any overt labeling
produces the same qualitative pattern of results. Experiment 2 demonstrates that labeling does not always
disrupt memory as predicted by the representational shift hypothesis: Differences in memory following
labeling versus preference are more likely an effect of judging preference, not an effect of overt labeling.
Labeling does not influence memory by shifting memory representations toward the category prototype.
Rather, labeling objects at the basic level produces memory representations that are simply less robust
than those produced by other kinds of study tasks.
Keywords: recognition memory, labels, categorization, depth of processing, representational change
Why do we give objects names? In addition to facilitating
communication, names exert a powerful influence on how we learn
about and understand objects and categories. For example, infants
categorize better when the same name is paired with objects from
the same category (Yoshida & Smith, 2005), even when these
objects are perceptually dissimilar (Plunkett, Hu, & Cohen, 2008).
Names also facilitate category learning for adults, even when
names are not actively used during the learning task (Lupyan,
Rakison, & McClelland, 2007).
Category names can also influence object perception, as in
categorical perception, where discrimination between two percep-
tually similar stimuli is easier if they cross a boundary determined
by linguistic categories (Gilbert, Regier, Kay, & Ivry, 2006; Rob-
erson & Davidoff, 2000). The discrimination advantage for cross-
category pairs is reduced when participants perform a concurrent
verbal, but not visual, task (Gilbert et al., 2006; Roberson &
Davidoff, 2000), implying that categorical perception depends on
the ability to use category names.
Categorical perception demonstrates a clear effect of having
names. What effects might using names have on object represen-
tations? Lupyan (2008) suggested that labeling objects systemati-
cally affects how objects are represented in memory and obtained
a particularly provocative result that we focus on here.
Lupyan (2008) briefly presented participants pictures of chairs
and lamps. In different blocks, they were asked to press a key
denoting the object’s name (“chair” vs. “lamp”) or their preference
(“like” vs. “don’t like”). During a surprise old–new recognition
memory test, participants were presented with studied items (old)
and lures (new) that differed only subtly from the study item (see
Figure 1). Recognition memory was lower for labeled objects than
for objects that were judged for preference. This difference in
recognition memory was driven by fewer hits for labeled objects,
without any difference in false alarms.
Lupyan (2008) explained these results with a representational
shift hypothesis: Overtly labeling an object activates features as-
sociated with prototypical examples from the object’s category. In
a top-down manner, these features become coactive with the visual
features of the perceived object, systematically altering the object
representation stored in long-term visual memory. Specifically,
overt labeling shifts the stored object representation toward the
category prototype. When that same object is viewed at test, its
representation no longer matches the shifted representation in
memory. That study–test mismatch leads to a false sense that the
old object is new, resulting in fewer hits.
The representational shift hypothesis is similar to the category
adjustment model (Huttenlocher, Hedges, & Vevea, 2000), accord-
ing to which representations become biased toward the center of
the category. In cases of inexact or incomplete representations,
This article was published Online First July 18, 2011.
Jennifer J. Richler, Isabel Gauthier, and Thomas J. Palmeri, Department
of Psychology, Vanderbilt University.
This article is based on a doctoral dissertation by Jennifer J. Richler
submitted to Vanderbilt University. This work was supported by the
Temporal Dynamics of Learning Center (SBE-0542013), a National Sci-
ence Foundation Science of Learning Center, and a grant from the James
S. McDonnell Foundation. We would like to thank Tim Curran, Sean
Polyn, Geoff Woodman, Mike Mack, and Matt Crump for insightful
comments and suggestions on this project, and Justin Barisich, Steph
Harrison, Sarah Muller, and Magen Speegle for assistance with data
Correspondence concerning this article should be addressed to Jennifer
J. Richler, Vanderbilt University, PMB 407817, 2301 Vanderbilt Place,
Nashville, TN 37240-7817. E-mail: email@example.com
Journal of Experimental Psychology:
Learning, Memory, and Cognition
2011, Vol. 37, No. 6, 1579–1587
© 2011 American Psychological Association
category information provides a meaningful basis for drawing
inferences about a category exemplar. In certain ways, the repre-
sentational shift hypothesis extends this model, suggesting that
overt labeling magnifies these prototype biases. Unlike the cate-
gory adjustment model, where category biases arise when objects
are reconstructed during memory retrieval (Crawford, Hutten-
locher, & Engerbretson, 2000), the representational shift hypoth-
esis posits that category information modifies how objects are
stored in visual long-term memory.
The representational shift hypothesis provides a provocative
explanation for the effects of labeling on object memory. More
broadly, the representational shift hypothesis has potentially im-
portant implications for understanding any cognitive processes that
rely on object representations stored in memory, potentially falsi-
fying many models of object recognition, memory, and categori-
zation. While some models have proposed interactive activation
between category names and object representations of the sort used
to explain representational shift (e.g., Rogers & Patterson 2007),
many others have assumed instead a largely feed-forward process
of object recognition and categorization (e.g., see Palmeri & Tarr,
In Experiment 1, we tested whether it is the act of overt object
labeling that is critical or whether representational shift is simply
a consequence of object categorization more generally. In Exper-
iment 2, we asked whether differences in memory following la-
beling versus preference can be explained by differences in the
quality or strength of memory traces created by those different
study tasks, without requiring that object representations be sys-
tematically shifted by labeling.
The representational shift hypothesis postulates an effect of
overt labeling on object representations. Lupyan (2008) argued
that this effect cannot be attributed to categorization on its own:
Any effect of categorization should lead to differences in false
alarms because only category-relevant features are encoded. The
representational shift hypothesis assumes that objects are encoded
the same way but that the stored visual memory representations are
distorted by activating the category label. This leads to a difference
in hits, not false alarms. Therefore, according to Lupyan (2008),
although categorization is certainly a component of labeling, it is
the overt act of labeling that produces representational shift, not
However, in Lupyan (2008), labeling and categorization were
confounded: Participants labeled objects as chairs or lamps by
pressing one of two response keys. Therefore, it is impossible to
link the memory effects uniquely to overt labeling and not to
categorization. Experiment 1 demonstrated that impaired recogni-
tion memory can arise from category-level processing on its own,
in the absence of any overt labeling or explicit categorization
response, and that this impairment is driven by a difference in hits,
not false alarms.
We eliminated explicit labeling by using a sequential matching
task with chairs and lamps. In different blocks, participants judged
whether or not two sequentially presented items were from the
same category (category matching) or were the exact same exem-
plar (exemplar matching). We hypothesized that memory for ob-
jects seen during category matching may be similar to effects of
labeling since only category-level information is relevant, whereas
memory for objects seen during exemplar matching may be similar
to effects of preference judgments since attention to details of the
object is required. Two objects were presented on every trial, and
whether the second object was a chair or lamp was randomized
with respect to whether the correct response was “same” or “dif-
ferent.” Consequently, participants were required to consider cat-
top two examples show chairs. The bottom two examples show lamps.
Paired lures might be a different color from the target, differ in the presence
or absence of a feature (e.g., armrests), or have a different height–width
ratio. (Note: The last pair of lamps differed in color). Adapted from
http://www.ikea.com. Copyright 1999–2011 by Inter IKEA Systems B.V.
Examples of target–lure pairs used in Experiments 1 and 2. The
RICHLER, GAUTHIER, AND PALMERI
egory membership during category-matching blocks without mak-
ing any explicit labeling or categorization response.
Both category-matching and exemplar-matching blocks con-
tained the same number of trials in which the same image was
presented consecutively and the correct response was “same.” For
both kinds of study blocks, memory was only tested for objects
presented on these “same” trials. Although we manipulated study
task, the only difference between test objects during recognition
was the task context in which these otherwise identical trials were
TN) undergraduates received course credit for participation. Data
from four participants were discarded for below-chance perfor-
Stimuli were 144 color pictures of chairs and lamps
from www.ikea.com. Each picture was 250 ? 250 pixels, showing
a single chair or lamp on a white background. There were 72 pairs
of chair and lamp pictures (36 per category), with each target
matched with a paired lure. Paired lures differed from targets in
small but noticeable ways (see Figure 1). Pictures were sorted into
four sets. Two sets contained 20 target–lure pairs and were des-
ignated target sets. Assignment as a target or lure was counterbal-
anced. Target objects were presented twice during the study phase,
with both presentations within the same trial. Two other filler sets
contained 16 object pairs. Both objects in each filler set pair were
shown once during the study phase and were used to create either
the category-matching or exemplar-matching context. One target
set and one filler set were assigned to the category-matching block
and another target and filler set to the exemplar-matching block for
each participant (counterbalanced).
On each matching trial, a fixation cross (500 ms)
was followed by the first image (300 ms), a random pattern mask
(500 ms), and the second image (300 ms). A question mark then
cued participants to respond. Participants had 700 ms to respond
and heard a tone if they responded too slowly, at which point the
trial timed out.
In the category-matching block, participants pressed 1 if the two
objects were from the same basic-level category or 2 if they were
from different basic-level categories. In the exemplar-matching
block, participants pressed 1 if the two objects were the exact same
object or 2 if they differed in any way. Participants completed one
exemplar-matching and one category-matching block (order coun-
terbalanced). A five-trial practice block where participants judged
if two sequentially presented tables were the same or different
shape preceded the experimental blocks.
There were 52 trials in each study block (see Figure 2 for trial
types and their frequency). In both blocks, there were 20 target
trials (created with targets from the target object set). On these
trials, the same image was presented consecutively, and the correct
response was “same.” The remaining 32 trials in each block were
created from objects in the filler sets and were designed to create
either a category-matching or exemplar-matching context. For
category-matching blocks, the remaining 32 trials consisted of 16
noncritical “same” trials, where the two objects were different
exemplars from the same category, and 16 “different” trials, where
the two objects were from different categories. For the exemplar-
Twenty-four Vanderbilt University (Nashville,
matching block, the remaining 32 trials consisted of matched
object pairs from the filler set and required a “different” response.
In this way, subtle differences needed to be detected on different
trials in the exemplar-matching block, and participants could not
rely on global similarity. Prior to the experiment, participants were
shown examples of “same” and “different” trials for each task.
After participants completed both matching blocks, a surprise
recognition memory test followed. They were informed that some
of the pictures would be old, exactly the same as those in the study
phase, and some would be new, differing only subtly in details
such as shape or color. Pictures were presented on the screen one
at a time, and participants were instructed to press 1 if the object
was “old” or 2 if the object was “new.” Pictures remained on the
screen until participants made a response. Recognition memory
was only tested for objects presented on target trials where the
same image was shown consecutively, requiring a “same” re-
sponse (from both category-matching and exemplar-matching
blocks); tested items only differed in the task context in which they
were presented. There were 80 test trials presented in a random
As shown in Figure 3, there were significantly more hits and
false alarms for objects in the exemplar-matching versus category-
matching block, hits: t(19) ? 4.66, p ? .001, d ? 1.51; false
alarms: t(19) ? 2.90, p ? .01, d ? 0.94. There was also a trend
toward higher overall recognition memory performance (d?) for
objects in the exemplar-matching than category-matching block,
t(19) ? 2.03, p ? .056, d ? 0.66, and correct response times (RTs;
see Table 1) were significantly faster for objects presented in the
exemplar-matching than category-matching block, t(19) ? 2.87,
p ? .01, d ? 0.93. In contrast to Lupyan (2008), we observed some
of the memory effect in RTs, not just d?.
Recognition memory was worse for objects studied in the con-
text of category matching than objects studied in the context of
and exemplar-matching (right) study blocks in Experiment 1. Both blocks
contain an equal number of target trials (shown in a box). Images adapted
from http://www.ikea.com. Adapted from http://www.ikea.com. Copyright
1999–2011 by Inter IKEA Systems B.V.
Trial types and their frequency in the category-matching (left)
LABELING AND VISUAL RECOGNITION MEMORY
exemplar-matching as reflected in longer RTs and marginally
lower d?. Our findings are qualitatively similar to differences
between labeling and preference (Lupyan, 2008), as hits were
higher for exemplar matching versus category matching, and show
that overt labeling of objects is not necessary to obtain the pattern
of results used to support representational shift; category-level
processing without any overt labeling is sufficient.
Lupyan (2008) characterized impaired recognition from labeling
as a direct result of overt labeling, not simply categorization.
Recognizing that overt labeling in his studies involved categori-
zation, he argued that actively using a category label has an
additional influence on subsequent memory above and beyond
categorization. Lupyan’s main argument was that the memory
effect is observed in hits, while any categorization effect should be
observed in false alarms. The false-memory literature suggests that
categorization leads to coarse encoding of category-relevant fea-
tures, resulting in false alarms to lures from the same category that
share these features (Koutstaal et al., 2003; Koutstaal & Schacter,
1997; Sloutsky & Fisher, 2004).
While we observed a small difference in false alarms in Exper-
iment 1, it was in the opposite direction from what the false-
memory literature would predict based on coarse encoding: False
memory suggests a higher false-alarm rate for objects studied
during category matching (Koutstaal & Schacter, 1997; Koutstaal
et al., 2003; Sloutsky & Fisher, 2004). We observed a higher
false-alarm rate for objects studied during exemplar-matching, not
category matching. Therefore, a categorization effect in memory is
not necessarily indexed by an increase in false alarms for catego-
rized items. Impaired memory for labeled objects can just as well
be explained by impaired memory for objects categorized at a
relatively coarse basic level, with or without any overt labeling.
The representational shift hypothesis suggests that overt label-
ing impairs recognition memory (Lupyan, 2008). However, an
alternative account is that preference judgments enhance recogni-
tion memory. Indeed, what makes Lupyan’s (2008) hypothesis so
intriguing is that one might expect labeling to have almost no
systematic influence as a study task because names are automati-
cally activated by objects all the time (Kikutani, Roberson, &
Hanley, 2010; Meyer & Damian, 2007; Morsella & Miozzo,
2002). Unfortunately, impairment for labeling versus enhancement
for preference cannot be distinguished in Lupyan because there
were only two tasks and no baseline.
Experiment 2 tested these competing possibilities. Partici-
pants made two judgments for each object on every study trial.
One group labeled all objects (primary labeling group). On
most trials, after labeling, participants also reported the location
of the image on the screen (i.e., above or below fixation). On a
small proportion of trials, after labeling, participants made a
preference judgment too. A second group of participants did the
converse. They made preference judgments for all objects (pri-
mary preference group) and then either made a location judg-
ment or labeled the object.
For all participants, some objects were given both labels and
preference judgments. Other objects were given labels and location
judgments (primary labeling group) or preference and location
judgments (primary preference group). Which judgment rules the
According to the representational shift hypothesis, memory will
be worse for any objects that are labeled at study, regardless of
whether labeling is accompanied by a location or preference judg-
ment. Labeling distorts representations stored in long-term visual
test for objects presented in the category-matching and exemplar-matching blocks in Experiment 1. Error bars
show 95% confidence intervals for the paired-sample t tests.
Overall performance (d?; Panel a) and hit and false-alarm rates (Panel b) on the recognition memory
Correct Response Times on the Recognition Memory Test for
Objects Presented in Each Study Task and Their Matched Lures
in Experiment 1
Study taskResponse times (ms)
RICHLER, GAUTHIER, AND PALMERI
memory, regardless of the other encoding task. The representa-
tional shift hypothesis predicts better memory for objects judged
for preference and location because those objects do not also
include overt labeling (see Figure 4, left). Any labeled objects
suffer from a representational shift, which produces a decrease in
hits during recognition.
An alternative account is that differences in memory following
preference versus labeling arise due to differences in memory
strength, without requiring that object representations are system-
atically shifted. According to this memory strength account, pref-
erence judgments might enhance memory, perhaps because judg-
ing preference is less automatic than labeling and requires more
effort. By this account, memory is equally good for any of the
three study-–task combinations that incorporate a preference judg-
ment, even when it is combined with labeling, but is relatively
worse for the combination of labeling and location (see Figure 4,
right). Note that we were not testing why preference judgments
enhance memory. Whatever the explanation, the pattern of results
predicted by a memory strength account is inconsistent with a
representational shift account.
versity community were given monetary compensation ($10) for
participation. Participants were randomly assigned to either the
primary labeling (n ? 12) or primary preference group.
Stimuli were 80 pictures of chairs and lamps (20
target–lure pairs per category) created in the same manner as
Experiment 1. Images were sorted into four sets (five target–lure
pairs per category). For each participant, one object set (counter-
balanced) was designated as the 25% second task set.
On each trial, participants saw a picture of a
chair or lamp presented above or below fixation (300 ms).
Participants in the primary labeling group were then probed to
Twenty-four members of the Vanderbilt Uni-
label the object, pressing one key for “chair” and another for
“lamp.” On 75% of trials, they were then probed to indicate the
location where the object was presented relative to fixation,
pressing one key for “above” and another for “below.” On 25%
of trials, following the labeling judgment, participants were
probed to make a preference judgment, pressing one key for
“like” and another for “dislike.” The two response probes were
presented sequentially, with each prompt remaining on the
screen until a response was made.
The procedure was identical for participants in the primary
preference group, except that their first response was always to rate
their preference for the object, they judged location on 75% of
trials, and they labeled the object on 25% of trials.
Participants in each group knew which judgment they would
always be making first (labeling or preference). Although they
were not informed of the exact proportion of location judgments
versus other second judgments (labeling or preference), they were
told that the location judgment would be probed more frequently.
The second task was probabilistic to ensure that participants were
not generating both the location and critical second responses
(labeling or preference) on every trial.
For each response type, one response was made with the left
hand and the other with the right hand. The same two keys were
used for all response types. The response probes were the words
“NAME?”, “RATE?”, or “PLACE?” printed in the center of the
screen for the labeling, preference, and location tasks, respectively.
The two response options were displayed on the bottom left and
right of the probe image.
Each object was presented twice during the study phase (once
above and once below fixation) for a total of 80 trials. The
primary judgment (e.g., labeling) was followed by a location
judgment on 60 trials, and the primary judgment (e.g., labeling)
was followed by the other judgment (e.g., preference) on 20
account (left) and the alternative memory strength account (right).
Predicted recognition memory performance in Experiment 2 based on the representational shift
LABELING AND VISUAL RECOGNITION MEMORY