Semantic information activated during retrieval contributes to later retention: Support for the mediator effectiveness hypothesis of the testing effect.
ABSTRACT Previous research has proposed that tests enhance retention more than do restudy opportunities because they promote the effectiveness of mediating information--that is, a word or concept that links a cue to a target (Pyc & Rawson, 2010). Although testing has been shown to promote retention of mediating information that participants were asked to generate, it is unknown what type of mediators are spontaneously activated during testing and how these contribute to later retention. In the current study, participants learned cue-target pairs through testing (e.g., Mother: _____) or restudying (e.g., Mother: Child) and were later tested on these items in addition to a never-before-presented item that was strongly associated with the cue (e.g., Father)--that is, the semantic mediator. Compared with participants who learned the items through restudying, those who learned the items through testing exhibited higher false alarm rates to semantic mediators on a final recognition test (Experiment 1) and were also more likely to recall the correct target from the semantic mediator on a final cued recall test (Experiment 2). These results support the mediator effectiveness hypothesis and demonstrate that semantically related information may be 1 type of natural mediator that is activated during testing.
- SourceAvailable from: Linda A. Henkel[Show abstract] [Hide abstract]
ABSTRACT: Background/Study Context: What people remember can be shaped by how they access and evaluate their memories as well as by events that happen after the original experience. This study examined how thinking about events in different ways after their occurrence can influence younger and older adults' memory for what really occurred. Methods: Younger adults (ages 18-22) and older adults (ages 65-88) saw and imagined pictures of objects and later evaluated each object 0, 1, or 3 times on a task that either required them to remember the objects in a general way (old or new?), in a more specific manner (perceived, imagined, or new?), or that required thinking about objects without regard to whether or how they were earlier experienced (e.g., judging their function or frequency in everyday life). Results: Results showed that probing items multiple times on the intervening tasks increased the number of items younger and older adults successfully remembered later but also increased source misattributions of claiming to have seen objects that were really imagined, with older adults showing lower recall but higher source errors. Exposure to items on the nonretrieval intervening tasks negatively affected later source memory, and remembering items without explicitly considering their source increased source errors even more that did the non-retrieval-based intervening tasks. Conclusions: These findings illustrate the negative impact of thinking about and nondiscriminately remembering past events on subsequent memory accuracy.Political Communication 10/2014; 40(5):555-577. · 1.24 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: The finding that trying, and failing, to predict the upcoming to-be-remembered response to a given cue can enhance later recall of that response, relative to studying the intact cue-response pair, is surprising, especially given that the standard paradigm (e.g., Kornell, Hays, & Bjork, 2009) involves allocating what would otherwise be study time to generating an error. In three experiments, we sought to eliminate two potential heuristics that participants might use to aid recall of correct responses on the final test and to explore the effects of interference both at an immediate and at a delayed test. In Experiment 1, by intermixing strongly associated to-be-remembered pairs with weakly associated pairs, we eliminated a potential heuristic participants can use on the final test in the standard version of the paradigm-namely, that really strong associates are incorrect responses. In Experiment 2, by rigging half of the participants' responses to be correct, we eliminated another potential heuristic-namely, that one's initial guesses are virtually always wrong. In Experiment 3, we examined whether participants' ability to remember-and discriminate between-their incorrect guesses and correct responses would be lost after a 48-h delay, when source memory should be reduced. Across all experiments, we continued to find a robust benefit of trying to guess to-be-learned responses, even when incorrect, versus studying intact cue-response pairs. The benefits of making incorrect guesses are not an artifact of the paradigm, nor are they limited to short retention intervals.Memory & Cognition 08/2014; · 1.92 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Evidence‐based teaching (EBT) entails the use of empirically validated pedagogical tools and techniques that promote student learning. We offer a rationale for why psychology instructors should embrace EBT in their classrooms. We then review five areas of evidence offering specific tools and techniques that improve learning and retention: the testing effect, spaced learning, metacognition, writing to learn, and interteaching. We then briefly discuss how three student self‐regulated choices can promote learning. Finally, we urge psychology teachers and students to use the discipline's experimental findings to enhance student learning.Australian Journal of Psychology 03/2013; 65(1). · 1.08 Impact Factor
Semantic Information Activated During Retrieval Contributes to Later
Retention: Support for the Mediator Effectiveness Hypothesis of the
Shana K. Carpenter
Iowa State University
Previous research has proposed that tests enhance retention more than do restudy opportunities because
they promote the effectiveness of mediating information—that is, a word or concept that links a cue to
a target (Pyc & Rawson, 2010). Although testing has been shown to promote retention of mediating
information that participants were asked to generate, it is unknown what type of mediators are
spontaneously activated during testing and how these contribute to later retention. In the current study,
participants learned cue–target pairs through testing (e.g., Mother: _____) or restudying (e.g., Mother:
Child) and were later tested on these items in addition to a never-before-presented item that was strongly
associated with the cue (e.g., Father)—that is, the semantic mediator. Compared with participants who
learned the items through restudying, those who learned the items through testing exhibited higher false
alarm rates to semantic mediators on a final recognition test (Experiment 1) and were also more likely
to recall the correct target from the semantic mediator on a final cued recall test (Experiment 2). These
results support the mediator effectiveness hypothesis and demonstrate that semantically related infor-
mation may be 1 type of natural mediator that is activated during testing.
Keywords: testing effect, retrieval practice, cued recall, mediator effectiveness
One of the more widely studied and robust phenomena in
current memory research is the testing effect—the finding that
taking a memory test on some information enhances later memory
for that information (see e.g., Roediger & Karpicke, 2006a). This
effect emerges even when testing is compared with an equivalent
number of opportunities to engage in further encoding of the
material (see e.g., Butler & Roediger, 2007; Carrier & Pashler,
1992) and when corrective feedback is not provided on the initial
test (see e.g., Carpenter & DeLosh, 2005, 2006). These findings
have led to the notion that there is something beneficial about the
retrieval process itself that contributes to later retention.
We currently know little about what this process involves,
however, and how it benefits retention. Toward developing a
theoretical explanation for the testing effect, Carpenter (2009)
recently proposed an elaborative retrieval hypothesis. According
to this view, recalling a target from a cue (e.g., Basket: _____) is
more likely to involve the activation of information that is related
to that cue (e.g., Eggs, Wicker) than is simply restudying the
cue–target pair (Basket: Bread). Activating these concepts is ben-
eficial because it provides extra information than can facilitate
recall of the target at a later time.
In support of this notion, a number of studies have shown that
the testing effect is greater under conditions in which the initial test
is made more difficult, for example by administering tests of recall
instead of recognition (see e.g., Kang, McDermott, & Roediger,
2007), administering the initial test after longer as opposed to
shorter time intervals (see e.g., Karpicke & Roediger, 2007; Pyc &
Rawson, 2009; Whitten & Bjork, 1977), and by testing a target
item (e.g., Bread) with a weakly related cue (e.g., Basket: _____)
as opposed to a strongly related cue (e.g., Toast: _____; Carpenter,
2009). Each of these manipulations renders the target information
less accessible at the time of an initial test, which would increase
the likelihood of activating this “extra” information that is bene-
ficial for retention.
One shortcoming of this hypothesis, however, is that it fails to
specify exactly what this extra information is. What type of infor-
mation is activated during initial retrieval that is helpful for later
retention? A recent article by Pyc and Rawson (2010) has begun to
address this important question. The authors propose a mediator
effectiveness hypothesis, in which tests are beneficial because they
are more likely than restudy opportunities to enhance the link
between a cue and target via mediating information (i.e., a word or
concept that links a cue to a target). Participants in this study
learned Swahili–English word pairs (e.g., Wingu: Cloud) either
through repeated restudying of the intact word pair or through an
equivalent number of attempts to recall the English word (Wingu:
_____) followed by restudy of the intact word pair. During re-
The author thanks Natalie Clark, Michael Clary, Riebana Sachs, Mat-
thew Erdman, Jessica Kloeppel, Kellie Olson, Jill Feipel, Sam White, and
Hana Ibrik for their assistance with data collection and scoring.
Correspondence concerning this article should be addressed to Shana K.
Carpenter, Department of Psychology, Iowa State University, W112 Lago-
marcino Hall, Ames, IA 50011-3180. E-mail: firstname.lastname@example.org
Journal of Experimental Psychology:
Learning, Memory, and Cognition
2011, Vol. ●●, No. ●, 000–000
© 2011 American Psychological Association
0278-7393/11/$12.00 DOI: 10.1037/a0024140
study, all participants were asked to generate a mediator that could
help them remember the target from the cue (e.g., Wingu 3
“Wing” 3 Cloud). On a later cued recall test, participants who
engaged in prior recall of the English word were not only better at
recalling the English word from the Swahili cue (i.e., the classic
testing effect) but were also better at recalling the mediators they
had previously generated, along with the association between the
mediator and the English word.
These results nicely demonstrate that tests with feedback pro-
mote memory of mediating information that participants were
asked to generate during learning. In usual studies of the testing
effect, however, participants are not specifically asked to generate
mediators. Instead, they are given a cue and simply asked to recall
the target. An important question, therefore, is how mediating
information might account for the testing effect that is observed
under normal circumstances.
The current study addressed this question by exploring what
type of information is spontaneously activated during an initial test
and how this contributes to later retention. Carpenter (2009) pro-
posed that one type of information activated during retrieval might
be semantic in nature. For example, the cue Basket: _____ may
activate concepts such as “Eggs,” “Flour,” or “Wicker.” If the
correct target is eventually retrieved (Bread), there is now a
structure of mediating semantic information that links Basket to
Bread. Although Carpenter’s results support this idea by demon-
strating a stronger testing effect for targets recalled from weakly
related cues (Basket: _____) compared with strongly related cues
(Toast: _____), memory for the supposed mediating information
(e.g., “Eggs,” “Flour”) was never tested.
The purpose of the current study was to directly measure the
information that is activated during retrieval and how this contrib-
utes to later retention. In two experiments, participants learned
weakly related cue–target pairs (e.g., Mother: Child) through ei-
ther testing (Mother: _____) or restudying (Mother: Child). In
Experiment 1 participants completed a final test of single-item
recognition that contained cues (e.g., Mother) and targets (e.g.,
Child), as well as new items that had never appeared before. Each
new item was either unrelated to any items that had appeared
before (e.g., Rabbit) or strongly related to one of the cues (e.g.,
Father). Because of its strong preexisting association with the cue,
this latter item is referred to as a semantic mediator. If semantic
mediators are more likely to be activated during an initial test
compared with during restudy, then participants who learn the
cue–target pairs through testing should exhibit higher false alarm
rates to the semantic mediators relative to participants who learned
the cue–target pairs through restudying.
In Experiment 2, final retention of the same target items was
measured through cued recall. This time, participants were asked
to recall the targets from the same cues as before (e.g., Mother:
_____), from a new cue that was weakly related to the target (e.g.,
Birth: _____), or from the semantic mediator (Father: _____). If
an initial cued recall test (e.g., Mother: _____) involves activation
of a semantic mediator (e.g., Father) during the process of retriev-
ing the correct target (Child), then the link between the mediator
and target (Father 3 Child) should be stronger for word pairs that
were learned through testing compared with through restudying.
Furthermore, this link should be stronger than the link between the
target and a new related cue (Birth 3 Child), because the latter
was not as likely to have been activated during initial retrieval.
to participate in order to fulfill partial requirements for an intro-
ductory psychology course at Iowa State University. Thirty par-
ticipants were randomly assigned to learn the word pairs through
testing, and 29 through restudying. These groups are hereafter
referred to as the test and study conditions, respectively.
Participants learned 16 cue–target pairs that had
an average associative strength of 3% according to the norms of
Nelson, McEvoy, and Schreiber (1998). These norms were also
used to obtain a semantic mediator (i.e., a word with a high
preexisting association with the cue). For example, when given the
cue Mother: _____, the word Father (semantic mediator) is gen-
erated 60% of the time. Each semantic mediator was associated
with its respective cue at 57%, on average, but was not associated
with the target member of the cue–target pair (e.g., Father bears no
preexisting semantic association with Child) or with any of the
other 15 targets. A complete list of the materials can be found in
In addition, 16 new items were obtained that were not related to
any of the cues, targets, or semantic mediators (see the unrelated
items in the Appendix). The recognition test thus contained 32 old
items (16 cues and 16 targets) and 32 new items (16 semantic
mediators and 16 unrelated items).
Design and procedure.
Participants first underwent an initial
encoding phase in which they saw each of the 16 cue–target pairs
one at a time and were asked to rate the relatedness between the
two words on a scale of 1 (not at all related) to 5 (highly related).
The cue and target appeared side-by-side in the center of the
computer screen. The cue appeared in a box on the left, and the
target appeared underlined in a box on the right. Participants were
asked to try to remember these words for an upcoming memory
Immediately after entering a rating between 1 and 5 for each
cue–target pair, participants completed a 20-s distractor task in
which they were asked to add together a series of single-digit
numbers appearing rapidly on the screen. Immediately after adding
the numbers together and entering a response to indicate the sum
total, participants in the test condition completed a test on the word
pairs, and participants in the study condition were shown the same
word pairs again.
Participants in the test condition saw the cue member of the pair
presented by itself on the left side of the screen, and they were
asked to type in the underlined word that had previously been
paired with it. Participants’ responses appeared on the right side of
the screen, and after pressing the Enter key, the program advanced
to the next item without providing feedback. Participants in the
study condition were shown each of the 16 word pairs again, along
with the same 1–5 rating scale, and were asked to rate the relat-
edness between the two words just as before. Each of the 16 word
pairs was presented once for both test and study conditions, in an
order that was randomized and different for each participant.
After all items had been presented as test or study, all partici-
pants engaged in a 5-min distractor task in which they were asked
to list as many of the U.S. states as they could. Participants were
Fifty-nine undergraduate students volunteered
shown a list numbered 1–50 on the computer screen, and they were
instructed to enter a state name next to each number.
At the end of 5 min, all participants were given the same
recognition test, which involved the 32 old items (16 cues and 16
targets) and the 32 new items (16 semantic mediators and 16
unrelated items) in a randomly presented order. Each of the 64
items was presented one at a time, in the center of the computer
screen, with a scale labeled 1 (I definitely did NOT see this word)
to 6 (I definitely saw this word). Participants responded by pressing
the 1–6 key for each item and were not given feedback about the
correctness of their responses.
Results and Discussion
Recognition test performance was calculated for both old and
new items as a function of whether words were learned through
test or study. First, the 6-point recognition scale was divided in
half such that a response of 4 or higher was considered a “yes.” All
analyses were then repeated, this time for those items that received
only a 5 or 6 (i.e., “yes” responses that were given with higher
confidence). Finally, all analyses were again repeated for those
items that received a 6 (i.e., “yes” responses given with the highest
Table 1 contains the hit rates for old items as a function of
learning condition (test vs. study) and confidence of the response.
Significant testing effects emerged regardless of confidence level.
Cues that were learned through test produced significantly higher
hit rates than did cues learned through study, and this was true
whether the hit rate was based on all “yes” responses, t(57) ? 2.65,
p ? .011; medium-confidence “yes” responses, t(57) ? 2.98, p ?
.005; or high-confidence “yes” responses, t(57) ? 2.51, p ? .016.
Hit rates for targets also revealed a significant testing effect
whether they were based on all “yes” responses, t(57) ? 1.97, p ?
.05 (two-tailed; p ? .02 one-tailed); medium-confidence “yes”
responses, t(57) ? 2.77, p ? .008; or high-confidence “yes”
responses, t(57) ? 2.61, p ? .013.
Table 2 contains false alarm rates to new items as a function of
learning condition (test vs. study) and the type of new item
(semantic mediator vs. unrelated). False alarm rates for semantic
mediators were higher for test than for study, whereas the opposite
pattern was observed for the unrelated items. A 2 ? 2 (Learning
Condition ? Item Type) mixed analysis of variance (ANOVA)
revealed that this interaction was significant for all “yes” re-
sponses, F(1, 57) ? 7.46, p ? .009, MSE ? .007; medium-
confidence “yes” responses, F(1, 57) ? 6.12, p ? .017, MSE ?
.006; and high-confidence “yes” responses, F(1, 57) ? 4.61, p ?
.037, MSE ? .005. All three analyses also revealed a main effect
for item type (all Fs ? 54, ps ? .001), such that the false alarm rate
was greater overall for semantic mediators than for unrelated
items. No main effect emerged for learning condition (all Fs ?
These results confirm the prediction that semantic mediators
(e.g., Father) are more likely to be activated as a result of recalling
a target from a cue (Mother: _____) compared with simply re-
studying the cue and target together (Mother: Child). Experiment
2 was conducted to explore whether semantic mediators are more
likely to be linked with targets (e.g., Father 3 Child) as a result
of testing compared with as a result of restudying.
Experiment 2 was identical to Experiment 1, in which half of the
participants learned the 16 cue–target pairs through test, and the
other half through study. This time, however, final retention was
tested via cued recall in which participants were asked to recall the
target (e.g., Child) from the same cue as before (e.g., Mother:
_____) or from a new cue that they had not seen before. The new
cues were either items that were weakly related to the target (e.g.,
Birth: _____) or semantic mediators that were not related to the
target (e.g., Father: _____).
If the process of recalling a target from a cue establishes a link
between a semantic mediator and a target, then recall of a target
from a semantic mediator cue should be higher for participants
who learned the cue–target pairs through test compared with
through study. Furthermore, participants who learned the cue–
target pairs through test should have an easier time recalling a
target from a semantic mediator cue (e.g., Father: _____) com-
pared with from a new related cue (e.g., Birth: _____), because the
latter was presumably less likely to have been activated and linked
with the target during the process of initial retrieval.
same participant pool as in Experiment 1. Of these participants, 31
Sixty-one participants were recruited from the
Hit Rates to Cues and Targets in Experiment 1 as a Function of
Learning Condition and Level of Confidence
All “yes” responses (4s, 5s, 6s)
Medium-confidence “yes” responses (5s, 6s)
High-confidence “yes” responses (6s only)
Standard deviations are given in parentheses.
False Alarm Rates to Semantic Mediators and Unrelated Items
in Experiment 1 as a Function of Learning Condition and Level
Level of confidenceTest Study
All “yes” responses (4s, 5s, 6s)
Medium-confidence “yes” responses (5s, 6s)
High-confidence “yes” responses (6s only)
Standard deviations are given in parentheses.
SEMANTIC MEDIATORS AND THE TESTING EFFECT
participants in the test condition were given the final test with same
cues, 10 with new related cues, and 10 with semantic mediator cues.
Ten participants in the study condition were given the final test with
same cues, 10 with new related cues, and 10 with semantic mediator
Materials, design, and procedure.
cue–target pairs from Experiment 1, 16 new cues were obtained
that had similar preexisting associative strength to the targets as
did the original cues (see the related items in the Appendix). The
new related cues were associated with only their respective targets
and not with any of the other 15 targets.
The learning phases for test and study were identical to those in
Experiment 1. In Experiment 2, however, participants were given the
final cued recall test after a 30-min retention interval during which
they worked on answering a variety of unrelated trivia questions.
At the beginning of the final test in Experiment 2, participants
in all conditions were informed that the cues might or might not be
the same ones that they originally saw. Participants were instructed
to read each cue and to report which one of the previously learned
underlined words came to mind when they saw that cue. Partici-
pants entered their responses by typing them onto the right-hand
side of the screen next to the cue that was presented. Each of the
16 cues appeared one at a time, in a different random order for
each participant. As in Experiment 1, the final test was self-paced
and feedback was not provided.
In addition to the 16
Results and Discussion
The proportion of targets correctly recalled on the final test was
calculated for both test and study conditions according to the three
types of cues that participants received (same cues, new related
cues, or semantic mediator cues). Results for the same cue condi-
tion revealed a classic testing effect, in that participants who
learned the word pairs through test recalled more targets (M ?
.960, SD ? .042) than did those who learned the word pairs
through study (M ? .900, SD ? .084), t(19) ? 2.101, p ? .05.
Figure 1 displays the average proportion of targets recalled on the
final test from independent cues (i.e., semantic mediator cues vs. new
related cues) for both test and study conditions. For participants in the
test condition, recall of targets from semantic mediator cues was
greater than from new related cues. The advantage of semantic me-
diator cues over new related cues was not as strong for participants in
the study condition, however. A 2 ? 2 (Learning Condition ? Cue
Type) between-subjects ANOVA revealed that this interaction was
significant, F(1, 36) ? 4.34, p ? .045, MSE ? .031.
A main effect also emerged for learning condition, F(1, 36) ?
16.91, p ? .001, MSE ? .031, indicating that targets were better
recalled from independent cues by participants in the test condition
(M ? .431, SD ? .248) compared with the study condition (M ?
.203, SD ? .139). Finally, a main effect emerged for type of cue,
F(1, 36) ? 9.60, p ? .005, indicating that targets were better
recalled overall from semantic mediator cues (M ? .403, SD ?
.249) than from new related cues (M ? .231, SD ? .176).
These results demonstrate that testing enhances performance on a
cued recall test, even when the final test cues are not the same ones
with which the targets were originally learned. Not all independent
had greater preexisting associative links to the targets than did seman-
tic mediator cues, the latter produced better final recall of targets. The
advantage of semantic mediator cues over new related cues occurred
to a much greater degree when targets were learned through test
compared with study. It appears, therefore, that semantic mediators
are more likely to be linked with targets as a result of initial retrieval
compared with as a result of restudy.
In two experiments, final retention of target items was better
when these items were learned through testing compared with
through restudying. This replicates a number of studies reporting
beneficial effects of testing (see e.g., Bjork, 1988; Carpenter,
Pashler, & Vul, 2006; Chan, McDermott, & Roediger, 2006; Cull,
2000; Izawa, 1992; Karpicke & Roediger, 2008; Kuo & Hirshman,
1996; Mcdaniel, Roediger, & McDermott, 2007). The current
article helps advance greater understanding of the causal mecha-
nisms underlying this phenomenon by proposing a theoretical
explanation for why tests are beneficial.
According to the mediator effectiveness hypothesis (Pyc &
Rawson, 2010), tests are beneficial because they are more likely
than restudy opportunities to promote the effectiveness of media-
tors (i.e., words or concepts that link a cue to a target). Pyc and
Rawson found that, supporting this view, participants who learned
Swahili–English word pairs (e.g., Wingu: Cloud) through testing
(Wingu: _____) compared with through restudying (Wingu:
Cloud) were more likely to remember a mediator that they had
generated to link the cue to the target (e.g., “Wing”) and were also
more likely to remember the target when cued with the mediator.
The current results address an important question that the mediator
effectiveness hypothesis does not, however. That is, how does medi-
ating information play a role in the typical testing effect, in which
participants are not specifically asked to generate mediators? Partic-
Semantic MediatorNew Related
Proportion of Targets Recalled
Type of Independent Cue
learning condition and type of independent cue. Participants learned cue–
target pairs (e.g., Mother: Child) through either test (Mother: _____) or
study (Mother: Child) and later were asked to recall the target from an
independent cue that had some preexisting association to the target—that
is, a new related cue (e.g., Birth: _____) or from an independent cue that
had no preexisting association with the target but was strongly associated
with the original cue—that is, a semantic mediator cue (e.g., Father:
_____). Relative to participants who learned the items through study, those
who learned the items through test showed a substantial advantage in
recalling targets from semantic mediator cues compared with from new
Final cued recall accuracy in Experiment 2, according to
ipants in the current study were never instructed to generate mediators
during learning but were simply asked to learn cue–target pairs
through either cued recall or restudy. In Experiment 1, participants
who learned these items through cued recall (e.g., Mother: _____)
were more likely to make false alarms to an item that was never
presented but closely associated with the cue—that is, a semantic
mediator (e.g., Father)—than were participants who learned the same
items through restudy (Mother: Child).
In Experiment 2, participants who learned the items through
testing (e.g., Mother: _____) were more likely to produce the
correct target (Child) on the final test in response to a semantic
mediator cue (e.g., Father: _____) compared with a new cue that
had some preexisting association with the target (e.g., Birth:
_____). Participants who learned the items through restudy, on the
other hand, did not demonstrate as large an advantage for semantic
mediator cues over new related cues. The link between a semantic
mediator and a target (e.g., Father 3 Child) is therefore more
likely to be strengthened through testing compared with through
The current results support the notion that retention is facilitated
by the presence of mediating information that links a cue to a
target. They also expand upon this notion by demonstrating direct
evidence of the type of mediating information that is spontane-
ously activated during retrieval—that is, when participants are not
specifically asked to generate mediators. It appears that retrieving
a target from a cue can activate a word or concept that is seman-
tically related to the cue, which then becomes linked with the
target and serves to facilitate its retrieval at a later time. Activation
of this mediator is less likely to occur during restudy, because the
availability of the answer reduces the need to activate additional
information in an attempt to retrieve the target.
This semantic mediator hypothesis provides a more specific
theoretical explanation for the testing effect than do previous
views based on elaborative retrieval (see e.g., Carpenter, 2009;
Carpenter & DeLosh, 2006; Glover, 1989). The semantic mediator
hypothesis proposes that, beyond the notion that tests are benefi-
cial because of extra information that is activated during retrieval,
this extra information can be a semantic word or concept that is
activated by the cue and becomes linked with the target during the
process of retrieving the target from the cue.
This hypothesis accounts for the testing effect itself, as well as
the tendency for the effect to be greater under conditions in which
initial retrieval is made more difficult (see e.g., Carpenter &
DeLosh, 2006; Kornell, Hays, & Bjork, 2009; Pyc & Rawson,
2009; Whitten & Bjork, 1977). Rendering target information less
accessible at the time of initial retrieval would presumably in-
crease the likelihood of activating semantically related information
that can mediate the link between a cue and target.
The hypothesis also accounts for the tendency of testing effects
to be greater when the final test is given after a longer, as opposed
to a shorter, time delay (see e.g., Roediger & Karpicke, 2006b;
Runquist, 1983). The organization of long-term memory is be-
lieved to be semantic in nature (see e.g., Bartlett, 1932), as con-
trasted with the more perceptual nature of short-term memory (see
e.g., Baddeley, 1976). If an initial test is beneficial because it
activates and links semantically related information from the cue
to the target, then the advantages of testing would be more likely
to show up on a final test that is delayed rather than immediate,
because the former is more sensitive to semantic processing. The
finding that testing reduces the rate of forgetting over time, there-
fore (see e.g., Carpenter, Pashler, Wixted, & Vul, 2008; Roediger
& Karpicke, 2006b; see also Kornell, Bjork, & Garcia, in press),
could be due to the tendency for tests to enhance semantic content
that is more likely to remain in long-term memory.
Finally, it is important to note that the semantic mediator hy-
pothesis provides an explanation for testing effects that have been
observed in verbal cued recall tasks. It is likely that this hypothesis
may not account for the effect that has been observed with a
variety of different types of tasks, such as those that involve spatial
processing (see e.g., see Carpenter & Pashler, 2007; Kang, 2010;
Rohrer, Taylor, & Sholar, 2010). It is hoped that future research
will explore potential boundary conditions of this and other theo-
retical accounts of the testing effect, as well as how best to use
tests to maximize learning and retention.
Baddeley, A. D. (1976). The psychology of memory. New York, NY: Basic
Bartlett, F. C. (1932). Remembering: A study in experimental and social
psychology. Cambridge, England: Cambridge University Press.
Bjork, R. A. (1988). Retrieval practice and the maintenance of knowledge.
In M. M. Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Practical
aspects of memory (pp. 397–401). New York, NY: Academic Press.
Butler, A. C., & Roediger, H. L., III. (2007). Testing improves long-term
retention in a simulated classroom setting. European Journal of Cogni-
tive Psychology, 19, 514–527. doi:10.1080/09541440701326097
Carpenter, S. K. (2009). Cue strength as a moderator of the testing effect:
The benefits of elaborative retrieval. Journal of Experimental Psychol-
ogy: Learning, Memory, and Cognition, 35, 1563–1569. doi:10.1037/
Carpenter, S. K., & DeLosh, E. L. (2005). Application of the testing and
spacing effects to name learning. Applied Cognitive Psychology, 19,
Carpenter, S. K., & DeLosh, E. L. (2006). Impoverished cue support
enhances subsequent retention: Support for the elaborative retrieval
explanation of the testing effect. Memory & Cognition, 34, 268–276.
Carpenter, S. K., & Pashler, H. (2007). Testing beyond words: Using tests
to enhance visuospatial map learning. Psychonomic Bulletin & Review,
14, 474–478. doi:10.3758/BF03194092
Carpenter, S. K., Pashler, H., & Vul, E. (2006). What types of learning are
enhanced by a cued recall test? Psychonomic Bulletin & Review, 13,
Carpenter, S. K., Pashler, H., Wixted, J. T., & Vul, E. (2008). The effects
of tests on learning and forgetting. Memory & Cognition, 36, 438–448.
Carrier, M., & Pashler, H. (1992). The influence of retrieval on retention.
Memory & Cognition, 20, 633–642. doi:10.3758/BF03202713
Chan, J. C. K., McDermott, K. B., & Roediger, H. L. III. (2006). Retrieval-
induced facilitation: Initially nontested material can benefit from prior
testing of related material. Journal of Experimental Psychology: Gen-
eral, 135, 553–571. doi:10.1037/0096-3418.104.22.1683
Cull, W. L. (2000). Untangling the benefits of multiple study opportunities and
repeated testing for cued recall. Applied Cognitive Psychology, 14, 215–
Glover, J. A. (1989). The “testing” phenomenon: Not gone but nearly
forgotten. Journal of Educational Psychology, 81, 392–399. doi:
Izawa, C. (1992). Test trial contributions to optimization of learning
processes: Study/test trials interactions. In A. F. Healy & S. M. Kosslyn
SEMANTIC MEDIATORS AND THE TESTING EFFECT