c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9

Available online at www.sciencedirect.com

Journal homepage: www.elsevier.com/locate/cortex

Research report

The generation effect: Activating broad neural circuits during

memory encoding

Zachary A. Rosner*, Jeremy A. Elman 1 and Arthur P. Shimamura 2

Department of Psychology, University of California, Berkeley, CA, USA

article info abstract

Article history: The generation effect is a robust memory phenomenon in which actively producing material
Received 18 April 2012 during encoding acts to improve later memory performance. In a functional magnetic
Reviewed 20 June 2012 resonance imaging (fMRI) analysis, we explored the neural basis of this effect. During
Revised 27 July 2012 encoding, participants generated synonyms from word-fragment cues (e.g., GARBAGE
Accepted 8 September 2012 eW_ST_) or read other synonym pairs (e.g., GARBAGEeWASTE). Compared to simply
Action editor Bradley Postle reading target words, generating target words significantly improved later recognition
Published online 21 September 2012 memory performance. During encoding, this benefit was associated with a broad neural
network that involved both prefrontal (inferior frontal gyrus, middle frontal gyrus) and
Keywords: posterior cortex (inferior temporal gyrus, lateral occipital cortex, parahippocampal gyrus,
Generation effect ventral posterior parietal cortex). These findings define the prefrontal-posterior cortical
Memory dynamics associated with the mnemonic benefits underlying the generation effect.
fMRI ª 2012 Elsevier Ltd. All rights reserved.
Encoding
Retrieval

1. Introduction phenomenon psychologists call the generation effect (Bertsch

et al., 2007; Slamecka and Graf, 1978). The benefits of gener-
Psychologists and educators have long extolled the impor- ation have been observed for many kinds of materials,
tance of mnemonic techniques for active learning, such as including verbal information (Bertsch et al., 2007; Slamecka
organizing material, monitoring learning, and practicing and Graf, 1978), arithmetic problems (Smith and Healy, 1998)
retrieval (Karpicke and Blunt, 2011; Metcalfe and Kornell, 2007; and pictures (Kinjo and Snodgrass, 2000). It has also been
Roediger and Karpicke, 2006). In one study (Karpicke and useful as a way of facilitating memory encoding in older
Roediger, 2008), individuals learned SwahilieEnglish word adults (Taconnat and Isingrini, 2004; Taconnat et al., 2006) and
pairs (e.g., mashua-boat) and practiced retrieving the English neurological patients (Lengenfelder et al., 2007; Souliez et al.,
words associated with the Swahili referents (e.g., mashua-?). 1996).
Compared to simply reading the word pairs, retrieval practice Psychological theories have suggested that the generation
significantly improved memory. Such findings demonstrate effect is driven by a host of internally mediated, topedown
the importance of self-generating informationda processes, such as conceptual analysis (Jacoby, 1983),

* Corresponding author. Department of Psychology, Room 3210 Tolman Hall, University of California, Berkeley, CA 94720-1650, USA.
E-mail addresses: [email protected] (Z.A. Rosner), [email protected] (J.A. Elman), [email protected] (A.P. Shimamura).
1
Tel./fax: þ1 510 643 5371.
2
Tel.: þ1 510 642 7131; fax: þ1 510 643 5371.
0010-9452/$ e see front matter ª 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.cortex.2012.09.009
1902 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9

semantic integration (McElroy, 1987), item distinctiveness conditions (see Paller and Wagner, 2002). In addition to the
(Begg et al., 1989; Kinoshita, 1989; Hunt and McDaniel, 1993), IFG, generation may increase activity in other areas also
and selective attention (Jurica and Shimamur, 1999; Tyler associated with this subsequent memory effect, including the
et al., 1979). Such processes may be defined more distinctly frontal operculum (FOP), fusiform gyrus (FG), inferior tempo-
by addressing the neural processes that drive the generation ral gyrus (ITG), cingulate gyrus, dorsal posterior parietal cortex
effect. Yet despite extensive behavioral analyses (for review (dPPC), and LOC (see Cansino et al., 2002; Kirchhoff et al., 2000;
see, Bertsch et al., 2007), no published study, to our knowl- Uncapher and Wagner, 2009; Wagner et al., 1998).
edge, has assessed the neural correlates of the generation In the present study, we employed a prototypical memory
effect. Candidate structures that could potentially drive this paradigm used to assess the generation effect. Participants
active encoding effect include those involved in topedown were shown related word pairs in the form of a cue word and
executive processing. For example, semantic retrieval and word fragment (e.g., QUARRELeF_GHT) and asked to complete
conceptual analysis, which lead to elaborative, long-lasting the second word in each pair. These encoding trials were
memory traces (Craik and Lockhart, 1972), have been linked compared to trials in which participants simply read related
to activity in the left inferior frontal gyrus (IFG) (Bookheimer, pairs (e.g., QUARRELeFIGHT) (Fig. 1A). At test, old/new recog-
2002; Baker et al., 2001; Poldrack et al., 1999). Other prefron- nition memory for the second word in each pair was assessed
tal regions, particularly in the dorsolateral prefrontal cortex with confidence ratings (high vs low) (Fig. 1B). Participants
(dlPFC), such as the middle frontal gyrus (MFG), have been were scanned during both study and test phases to identify
associated with other executive control processes presumed the neural substrates underlying the generation effect.
to interact dynamically with posterior regions (see Miller and
Cohen, 2001; Shimamura, 2000, 2008). For example, dlPFC re-
gions have been associated with a variety of working memory 2. Materials and methods
processes that lead to long-term memory formation (Paller
and Wagner, 2002), such as refreshing perceptual features, 2.1. Participants
maintaining items in memory, manipulating information, and
selecting items for retrieval (Cohen et al., 1997; D’Esposito Twenty-four healthy individuals (13 females, 11 males, mean
et al., 1997, 1999; Johnson et al., 2005; Postle, 2006; Raye age ¼ 23 years, range ¼ 18e32 years, all right-handed, native
et al., 2002; Thompson-Schill et al., 1997). English speakers) participated in the study. Informed consent
To the extent that the generation effect is mediated by item was obtained according to guidelines approved by the UC
distinctiveness, it may be that posterior regions involved in Berkeley Office for the Protection of Human Subjects. No
verbal or item analysis, such as the left middle temporal gyrus participants reported any history of neuropsychiatric disorder
(MTG) and lateral occipital cortex (LOC) (Binder et al., 2009; or recent use of psychoactive medication. Participants were
Cabeza and Nyberg, 2000; Malach et al., 1995) also become compensated $12 per hour.
particularly involved. Additionally, one might predict
increased activation in the anterior cingulate cortex (ACC), 2.2. Stimuli
which is involved in conflict monitoring (van Veen et al., 2001)
and verbal generation (Barch et al., 2000). Finally, with respect A total of 200 cue-target synonym word pairs were con-
to monitoring internally or cognitively mediated processing, structed (e.g., GARBAGEeWASTE). One hundred items were
the generation effect may map onto activation related to the presented at study and again at test, while the other 100 items
so-called default mode network (DMN), initially observed were used as lures at test. Target words were obtained from
during periods of “rest”, such as between stimulus pre- the MRC Psycholinguistic Database (Wilson, 1988) and con-
sentations (Raichle et al., 2001). The DMN is a set of brain re- sisted of a mean word length of 5.39 letters (range ¼ 3e8 let-
gions that include the dorsal medial prefrontal cortex ters), and a mean frequency of 54.32 (range 1e314) (Francis
(dMPFC), ventral medial prefrontal cortex (vMPFC), posterior and Kuc era, 1982). During encoding, target words were pre-
cingulate cortex (PCC), inferior parietal lobule (IPL), precuneus sented in fragmented form (generate condition; e.g., GARBA-
(PrC), retrosplenial cortex (Rsp), lateral temporal cortex (LTC), GEeW_ST_) or in complete form (read condition; e.g.,
and hippocampal formation. Upon further analysis, this GARBAGEeWASTE). Fragments were created by removing
network has been associated with various internally mediated each vowel (unless it began a word) and replacing it with an
processes, such as episodic recollection, prospective memory, underline score. The encoding strategy (read vs generate) and
and perspective taking (see Buckner et al., 2008; Buckner and mnemonic status (old vs new) of each word were counter-
Carroll, 2007; Spreng et al., 2009). Given the view that the balanced across participants.
generation effect is involved in internally mediated process-
ing, one might expect greater DMN activation during encoding 2.3. Behavioral procedure
for generate versus read items.
With respect to long-term memory processes, activity in The study phase was presented in two separate scanning
the IFG during encoding has been particularly associated with blocks, each consisting of a randomized presentation of 25
successful retrieval (Brewer et al., 1998; Wagner et al., 1998; for generate and 25 read trials. For each study trial, the stimulus
review see Paller and Wagner, 2002). Specifically, the IFG is (either intact or fragmented pairs) was shown for 3 sec which
more active during encoding for items subsequently remem- was followed by a 500 msec blank screen and a jittered fixation
bered compared to those subsequently forgotten. This effect is cross (4e8 sec). Participants were instructed to make a key-
robust and has been observed in a variety of tasks and press response when they could identify the second word in
 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9 1903

Fig. 1 e Experimental design and behavioral data. (A) Experimental design of encoding phase and (B) retrieval phase. (C)
Recognition accuracy for read and generate items. Hits are items correctly identified as old. HC hits are items correctly
identified as old with HC.

each pair (i.e., the target word). This procedure encouraged (TE) ¼ 22 msec, flip angle ¼ 90 , matrix ¼ 128  128,
comparable processing across study conditions, except that FOV ¼ 220 mm, 1.7  1.7 in-plane resolution) with GRAPPA
fragmented items had to be generated (Fig. 1A). (acceleration factor 3). For functional scans, EPIs consisted of 37
Following the study set, a 3-min filled retention interval axial slices, 2.5 mm thick, oriented to the anterior–posterior
was presented. During this interval, participants were shown commissure (AC–PC), and were acquired in an interleaved order
24 simple math equations (e.g., 3 þ 5 ¼ 8) and determined which resulted in whole brain coverage. A total of 155 volumes
whether the answer was true or false. Thereafter, old/new (run duration ¼ 310 sec) were collected during each of two
recognition memory was assessed using the 50 target items encoding runs and 255 volumes (run duration ¼ 510 sec) were
and 50 new items. New items were target words from unused collected during each of two retrieval runs. The first five vol-
word pairs. For each test trial, a word was presented for umes of each run were used for magnetization preparation and
500 msec, followed by a 3 sec blank screen, and a jittered were removed from future analyses, resulting in 150 and 250
response interval (4e8 sec) (Fig. 1B). Participants determined volumes for each encoding and retrieval session, respectively.
whether a test word was old or new while indicating their For registration purposes, a high resolution magnetization-
confidence (high or low) for each response during the intertrial prepared rapid-acquisition gradient echo (MPRAGE) volume
interval (ITI). They were instructed to respond old with high (TR ¼ 2300 msec, TE ¼ 2.98, matrix ¼ 256  256, FOV ¼ 256,
confidence (HC) only if they were absolutely certain that the sagittal plane, slice thickness ¼ 1 mm, 160 slices) and a gradient-
test item was presented during the study phase. Thus, we echo multislice (GEMS) volume (TR ¼ 250 msec, TE ¼ 3,
interpret such HC hits to reflect strong recollective responses. matrix ¼ 256  256, FOV ¼ 220, 3 mm slice thickness, 28 slices)
Upon completion of the first study-test block, the behavioral were collected. Due to movement artifacts, eight of the 96 runs
procedure was repeated with a different set of cue-target pairs. were excluded from data analysis.

2.4. fMRI acquisition 2.5. fMRI data analysis

A 3T Siemens (Erlangen, Germany) Trio scanner housed at the All data processing and analyses were performed using the
UC Berkeley Brain Imaging Center was used to acquire T1- FMRIB Software Library (FSL) toolbox v4.1.4 (http://www.
weighted anatomical images and T2*-weighted echo-planar fmrib.ox.ac.uk/fsl; Smith et al., 2004). During preprocessing,
images (EPIs) (repetition time (TR) ¼ 2000 msec, echo time brain extraction tool (BET) was applied to each participant’s
1904 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9

data to separate brain tissue from skull and dura using a mask that was 22% greater than that for read items [generate
of the brain from the first volume, which was used for sub- hits ¼ 87%, read hits ¼ 65%, t(23) ¼ 9.97, p < .001, false alarm
sequent volumes. Images were then spatially smoothed using rate ¼ 21%; see Fig. 1C, Table 1]. The difference between the
a 5 mm full width at half maximum (FWHM) of the Gaussian two conditions was even greater when performance was
kernel. To remove low frequency artifacts, highpass temporal based only on high-confident hits [generate HC hits ¼ 74%,
filtering was performed with the local Gaussian-weighted fit read HC hits ¼ 42%, t(23) ¼ 11.61, p < .001, HC false alarm
of a running line. Motion Correction using FMRIB’s Linear rate ¼ 7%; see Fig. 1C, Table 1]. As mentioned above, an HC
Image Registration Tool (MCFLIRT) corrected for motion by rating was made when participants were absolutely certain that
aligning images to the middle slice with rigid body trans- they had seen a test item during the study phase. Given our
formation. Sinc interpolation (Hanning windowed) shifted findings for HC hits, we can assert that the generation effect is
each slice in the volume in reference to the middle of the TR particularly potent in driving strong recollective responses.
period. Next, FMRIB’s Linear Image Registration Tool (FLIRT) During encoding, the ability to identity targets was high and
registered subject’s EPIs to their skull-stripped high resolution not significantly different between generated and read targets
T1-weighted images, which were then registered to standard [generated targets ¼ 98%, read targets ¼ 99%, t(23) ¼ 1.89,
Montreal Neurological Institute space (FSL’s MNI152 tem- p ¼ .07]. Mean latency to identify a target was longer for
plate), both of which were combined to transform the EPI’s generated items than read items [generate ¼ 843 msec,
and statistical maps into standard space. read ¼ 670 msec, t(23) ¼ 6.58, p < .001].
At the first level of analysis, a multilevel, mixed effects
general linear model was run using FMRIB’s Improved Linear 3.2. fMRI data
Model (FILM). Each individual run (two encoding and two
retrieval runs per participant) was modeled in individual We first assessed memory-related effects by contrasting ac-
subject space. Next, each resulting statistical map was regis- tivations during encoding for items that were subsequently
tered to standard space. Regressors of interest were obtained remembered with those that were subsequently forgotten,
by convolving stimulus onset times with FSL’s canonical collapsed across encoding condition. This contrast revealed
(gamma) hemodynamic response function and temporal de- significant activation in the left LOC. In a second analysis, we
rivative. Trials in which the participant failed to respond, assessed items that elicited HC (i.e., strongly recollected) rat-
including those trials in which the participant was unable to ings. This contrast revealed significant activation in the left
identify the target word at encoding, were included in the LOC, IFG, ITG, and right precentral gyrus. Thus, with respect to
model as regressors of no interest. Finally, motion parameters encoding effects, memory-related activity was particularly
were added as a confound variable and temporal autocorre- observed for items remembered with HC (i.e., strong
lation was removed through prewhitening. recollections).
At the second level of analysis, each subject’s two encoding We were particularly interested in determining the neural
runs were combined, as were each subject’s two retrieval runs, processes that drive contrasts between generate and read
using one-sample t-tests. These runs were treated as fixed ef- conditions. We thus assessed the contrast of generate
fects. At the third level, statistical maps were created at the hits > read hits, which resulted in significant activation in IFG,
group level for each contrast using FMRIB’s Local Analysis of MFG, LOC, PrC, ITG, intraparietal sulcus (IPS), and ACC (see
multilevel GLM Mixed Effects (FLAME). The whole-brain fam- Fig. 2A, Table 2). The reverse contrast (read hits > generate
ily-wise error was corrected to p < .05 using Gaussian Random hits) resulted in no significant activation differences. We next
Field theory with a cluster forming threshold of Z > 2.3. To assessed neural activations associated with the generation
assess the relationship between behavioral performance and effect for HC hits (generate > read, HC hits), which revealed
neural activity, we applied two separate subject-specific activations in bilateral IFG, MFG, LOC, ITG, IPS, PrC and ACC
covariate analyses. First, we used individual generation effect (see Fig. 2B, regions in red, Table 3). The reverse contrast
recollection benefit (generate HC hit rateeread HC hit rate) as a (read > generate, HC hits) revealed activation in bilateral LOC
covariate of interest in an analysis of generation effect recol- and PrC, and left angular gyrus (AG) (see Fig. 2B, regions in
lection activity (generate > read, HC hits). We used HC blue, Table 3).
responses to isolate recollection responses and eliminate the Activations during retrieval were consistent with previous
confound of varying memory strength and remove possible findings of the successful retrieval effect in which hits are
guess trials. In addition, we used individual memory perfor- compared with correction rejections (CR) (hits > CRs). In the
mance (hit-false alarm score) as a covariate of interest in an present study, the successful retrieval effect was associated
analysis of overall generation effect activity (generate > read, with increased activation in the left IFG, MFG, superior frontal
all items). Localizations of peak activations were identified by gyrus (SFG), LTC, LOC, ACC, supramarginal gyrus (SMG), and
mapping images onto the Harvard-Oxford Cortical Atlas. AG (Fig. 3A). This retrieval-based network was observed when

3. Results Table 1 e Recognition accuracy for generate and read

conditions.
3.1. Behavioral performance Hit False alarm HC hit HC false alarm

Generate .87 .21 .74 .07

We confirmed the robust benefit afforded by the generation
Read .65 .21 .42 .07
effect. Specifically, the generate condition produced a hit rate
 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9 1905

Fig. 2 e Statistical activation maps for the generation effect during encoding. (A) Hits. Generate > read (red): regional
activations include bilateral IFG, MFG, LOC, PrC, ITG, IPS, ACC. Read > generate (blue): no significant activation. (B) HC hits.
Generate > read (red): regional activations include bilateral IFG, MFG, LOC, ITG, IPS, ACC, right PrC. Read > generate (blue):
bilateral LOC, PrC, left AG.

contrasts were restricted to generated items (hits > CRs, hits during encoding. With this analysis, we addressed the
generate items, Fig. 3B) or to read items (hits > CRs, read items, degree to which overall memory performance may be medi-
Fig. 3C). Direct comparisons of retrieval-based generated ated by the magnitude of neural activations associated with
versus read items revealed no reliable differences. the generation effect across individuals. As shown in Fig. 4A,
To evaluate neural correlates of the generation effect with memory performance was significantly correlated with ac-
respect to behavioral performance, we performed a covariate tivity in the right parahippocampal gyrus (PHG), temporal
analysis of recognition performance and regional neural re- fusiform cortex, MTG, AG, LOC, and PrC. Thus, the strength of
sponses associated with the generation effect. We used as our activation within these regions elicited by self-generation at
covariate of interest overall memory performance (hit-false encoding predicted better memory performance. As the gen-
alarm) and correlated it with the contrast of generate > read eration effect was particularly potent for HC hits, we per-
formed a second covariate analysis in which the behavioral
advantage of generation for HC hits (generate HC hit ratee-
read HC hit rate) was correlated with its neural counterpart,
Table 2 e Brain regions active at encoding for subsequently the contrast of generate HC hits > read HC hits. In this anal-
remembered items. Generate hit > read hit; Read
ysis, we found correlated activity in the paracingulate, frontal
hit > generate hit (MNI coordinates).
pole, left ACC, and right SFG (Fig. 4B), suggesting a medial-
Cluster index BA Z x y z frontal network underlying the behavioral benefit of genera-
tion for producing strong recollective responses (i.e., HC hits).
Generate hit > read hit
Anterior cingulate gyrus 24 4.73 0 0 28
Left IFG 48 5.44 52 14 28
Right IFG 45 4.42 50 34 16
Left ITG 37 4.34 46 56 16 4. Discussion
Right ITG 37 4.70 46 54 10
Right insular cortex 48 3.10 32 0 10 The present findings addressed the neural correlates of the
Left LOC 19 5.98 24 76 26
generation effect. Active generation was associated with a
Right LOC 19 6.13 28 76 32
broad set of regions that included the IFG, MFG, ACC, PrC, IPS,
Left MFG 44 4.78 54 14 34
Right MFG 45 4.50 52 24 26 ITG, and LOC. Significant prefrontal activity (IFG and MFG)
Left occipital pole 18 5.56 32 90 8 confirmed the role of executive control processes important
Right occipital pole 18 5.71 34 92 2 for establishing long-term memories. Thus, these findings
Left precentral gyrus 44 5.95 44 4 24 mesh well with studies that have shown that these regions are
Right precentral gyrus 48 5.17 42 8 26 particularly involved in stimulus refreshing, updating, and
Left PrC 7 3.23 12 62 42
semantic access (D’Esposito et al., 1997; Johnson et al., 2005;
Right PrC 7 3.72 18 58 42
Raye et al., 2002; Thompson-Schill et al., 1997). For example,
Left superior parietal lobule 40 4.92 38 42 44
Right superior parietal lobule 7 4.45 28 56 48 previous studies have shown that these PFC regions are active
when participants must refresh or re-activate recently pre-
Read hit > generate hit
sented words, drawings, or patterns (Johnson et al., 2005; Raye
No significant activations
et al., 2002). The generation effect can thus be linked to related
1906 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9

monitoring and control (D’Esposito et al., 1999; Miller and

Table 3 e Brain regions active at encoding for items
subsequently remembered with HC. HC generate > HC Cohen, 2001; Postle, 2006; Shimamura, 2008).
read; HC read > HC generate (MNI coordinates). Importantly, covariate analyses showed that memory
performance could be predicted by the degree to which neural
Cluster index BA Z x y z
networks associated with the generation effect were active.
HC generate > HC read Specifically, we found that overall memory performance was
Left central opercular cortex 48 4.04 46 6 2 correlated with increased generate activity in the PHG, tem-
Left cingulate gyrus 32 3.73 4 40 18 poral fusiform cortex, MTG, AG, LOC, and PrC. In addition, the
Right cingulate gyrus 24 3.15 4 30 22
behavioral benefit of generating at encoding to produce sub-
Left frontal orbital cortex 47 4.14 42 26 6
Left frontal pole 47 3.42 42 38 14
sequent HC hits was correlated with activity in medial ante-
Left IFG 45 5.25 48 34 6 rior PFC regions known to be important for attending to
Right IFG 48 3.08 42 16 24 internally generated versus externally perceived stimuli
Left ITG 37 3.96 58 60 20 (Lagioia et al., 2011; Simons et al., 2006, 2008). These findings
Right ITG 37 5.7 54 56 14 link the generation effect to regional activations during
Left insular cortex 48 4.38 32 18 6
encoding that are known to be critical for the establishment of
Left LOC 19 6.07 24 76 28
long-term memories (Paller and Wagner, 2002). Generation
Right LOC 19 5.58 32 72 26
Right lingual gyrus 18 3.32 10 88 6 increased both prefrontal activity and activity in posterior
Left MFG 45 3.99 44 30 26 regions involved in verbal processing, object analysis, and
Right MFG 6 3.27 54 10 42 visuospatial imagery. Additionally, participants who benefited
Left occipital FG 19 3.67 40 74 20 the most from generation showed the greatest activation in
Right occipital FG 19 3.47 30 80 4 regions known to be important for memory binding and
Left occipital pole 18 6.11 34 94 10
retrieval, such as the PHG, AG, and PrC (see Davachi, 2006).
Right occipital pole 18 5.5 34 92 2
Recently, Moss et al. (2011) compared activation when
Left paracingulate gyrus 32 3.35 4 16 48
Right paracingulate gyrus 32 3.67 12 32 28 participants reread, paraphrased, or explained biology texts.
Right postcentral gyrus 2 4.54 48 28 48 While self-explaining led to the greatest memory benefit,
Right postcentral gyrus 3 3.22 56 20 46 regional activity in ACC, bilateral superior parietal cortex, and
Left precentral gyrus 44 6.06 48 6 26 left IFG also increased along with complexity of semantic
Right precentral gyrus 44 4.78 44 8 26 processing. In the present study, different regions within the
Left PrC 18 3.39 16 70 30
DMN were active when reading or generating items during
Right PrC 19 3.35 24 64 30
Left SFG 32 3.99 6 38 42
encoding (IPL, PrC, dMPFC for generate > read, HC hits; IPL, PrC
Left SFG 8 3.6 4 28 48 for read > generate, HC hits), suggesting that the DMN is
Left superior parietal lobule 40 4.95 36 48 46 responsible for internally driven processing, though different
Right superior parietal lobule 7 4.6 32 54 48 regions may mediate different topedown processes. It is
Left SMG 40 4.67 44 44 46 possible that on some trials, active generation oriented par-
Right SMG 2 3.81 44 36 48
ticipants to internally generated information arising from
Left temporal occipital fusiform cortex 37 4.39 44 56 16
semantic analysis or conceptual processes, while reading kept
Right temporal occipital fusiform cortex 37 5.04 44 58 12
participants less on-task and allowed for increased mind
HC read > HC generate wandering. It is acknowledged that the DMN is associated
Left AG 39 3.37 48 68 26
with many internally mediate processes and that there may
Left LOC 19 3.38 40 84 36
Right LOC 7 3.23 8 68 36
be regional specificity within the network depending on the
Left occipital pole 19 3.66 42 68 36 particular process being engaged (Buckner and Carroll, 2007;
Left PrC 23 3.43 10 68 26 Shimamura, 2011; Spreng et al., 2009).
Right PrC 7 3.23 8 68 36 At retrieval, successful recognition (hits > CRs) was asso-
Right superior parietal lobule 7 3.06 4 64 48 ciated with activation in lateral and medial PPC, two regions
associated with memory recollection (Cabeza, 2008;
Shimamura, 2011; Vilberg and Rugg, 2008). Interestingly, this
pattern of activity was observed for both successfully
acts of refreshing and updating, which also involve internally retrieved generated and read items, and there were no dif-
mediated or generated information. ferences during retrieval that differentiated remembered
As suggested by theories of executive control (Miller and items between the two conditions. Within the confines of the
Cohen, 2001; Shimamura, 2000, 2008), prefrontal mecha- encoding conditions used in the study, our findings suggest
nisms act to modulate or control posterior cortical activity that a remembered item (hit or high-confident hit) elicits the
thus engaging a broad prefrontal-posterior network involved same pattern of activation during retrieval regardless of
in selecting, maintaining, and manipulating information in whether it was previously generated or read.
working memory. In the present study, generation was asso- With respect to mapping psychological theories of the
ciated with both PFC and posterior activity, particularly in generation effect onto our functional magnetic resonance
regions involved in image generation (ITG) and object pro- imaging (fMRI) findings, it is clear that multiple brain regions
cessing (LOC) (see D’Esposito et al., 1997; Malach et al., 1995). are responsible for different aspects of the mnemonic benefit
Thus, the generation effect offers a useful analysis of the associated with the generation effect. Certainly, PFC regions
neural dynamics associated with executive or metacognitive involved with semantic analysis, refreshing, and updating are
 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9 1907

Fig. 3 e Statistical activation maps during retrieval. (A) Overall hits > correct rejections. Regional activations include left IFG,
MFG, SFG, ITG, MTG, LOC, ACC, SMG, AG. (B) Generate hits > correct rejections. Regional activations include left LOC, ACC,
SMG, AG. (C) Read hits > correct rejections. Regional activations include left IFG, MFG, SFG, ITG, MTG, ACC, SMG, AG, PHG.

Fig. 4 e Covariate analyses. (A) Shown in red are regions related to the generation effect (generate > read, all items) that
covaried with overall memory performance (hits-false alarms). Regional activations include PHG, MTG, AG, LOC, temporal
fusiform cortex, PrC. (B) Shown in red are regions related to the generation effect (generate > read, HC items) that covaried
with the behavioral generation effect (HC hits-false alarms). Regional activations include bilateral paracingulate cortex and
frontal pole, left ACC, and right SFG.

involved in driving the mnemonic benefit associated with the network of both PFC and posterior regions during encoding
generation effect. However, a host of posterior regions, such (see Shimamura, 2010).
as the PHG, temporal fusiform cortex, MTG, AG, and LOC, is
also involved. It is possible that active generation increases
attention and cognitive effort (prefrontal and posterior
cortical activation; Miller and Cohen, 2001; Shimamura, 2000, Acknowledgments
2008), conceptual and semantic processing (IFG and MTG;
Bookheimer, 2002; Poldrack et al., 1999), and item distinc- Supported by NSF Grant BCS-0745835 and NIH Grant
tiveness (LOC and ACC; Malach et al., 1995; van Veen et al., NS040813. We thank E. Klostermann Wallace, I. Chen,
2001). Perhaps one of the reasons memory researchers have L. Brown, B. Cohen-Sheehy, M. Park, and R. Stevenson for their
not reached a consensus regarding the underlying mecha- contributions in data collection, analyses, and comments.
nism of the generation effect is that active generation engages
a large range of cognitive processes. Depending on the task at
hand, active generation may promote increases in attention, references
cognitive effort, item distinctiveness, semantic processing,
and conceptual processing. Indeed, our findings affirm the
fact that these memory-related influences associated partic- Baker JT, Sanders AL, Maccotta L, and Buckner RL. Neural
ularly with strong recollective responses are driven by a broad correlates of verbal memory encoding during semantic and
1908 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9

structural processing tasks. NeuroReport, 12(6): 1251e1256, Karpicke JD and Blunt JB. Retrieval practice produces more
2001. learning than elaborative studying with concept mapping.
Barch DM, Brave TS, Sabb FW, and Noll DC. Anterior cingulate Science, 331(6018): 772e775, 2011.
and the monitoring of response conflict: Evidence from an Karpicke JD and Roediger HL. The critical importance of retrieval
fMRI study of overt verb generation. Journal of Cognitive for learning. Science, 319(5865): 966e968, 2008.
Neuroscience, 12(2): 298e309, 2000. Kinjo H and Snodgrass JG. Does the generation effect occur for
Begg I, Snider A, Foley F, and Goddard R. The generation effect is pictures? American Journal of Psychology, 113(1): 95e121, 2000.
no artifact: Generating makes words distinctive. Journal of Kinoshita S. Generation enhances semantic processing? The role
Experimental Psychology: Human Learning and Memory, 15(5): of distinctiveness in the generation effect. Memory and
977e989, 1989. Cognition, 17(5): 563e571, 1989.
Bertsch S, Pesta BJ, Wiscott R, and McDaniel MA. The generation Kirchhoff BA, Wagner AD, Maril A, and Stern CE. Prefrontal-
effect: A meta-analytic review. Memory and Cognition, 35(2): temporal circuitry for episodic encoding and subsequent
201e210, 2007. memory. The Journal of Neuroscience, 20(16): 6173e6180, 2000.
Binder JR, Desai RH, Graves WW, and Conant LL. Where is the Lagioia A, Eliez S, Schneider M, Simons JS, Van der Linden M, and
semantic system? A review and meta-analysis of 120 Debbané M. Neural correlates of reality monitoring during
functional neuroimaging systems. Cerebral Cortex, 19(12): adolescence. NeuroImage, 55(3): 1393e1400, 2011.
2767e2796, 2009. Lengenfelder J, Chiaravalloti ND, and DeLuca J. The efficacy of the
Bookheimer S. Functional MRI of language: New approaches to generation effect in improving new learning in persons with
understanding the cortical organization of semantic traumatic brain injury. Rehabilitation Psychology, 52(3): 290e296,
processing. Annual Review of Neuroscience, 25: 151e188, 2002. 2007.
Brewer JB, Zhao A, Desmond JE, Glover GH, and Gabrieli JDE. Making Malach R, Reppas JB, Benson RR, Kwong KK, Jlan H,
memories: Brain activity that predicts how well visual experience Kennedy WA, et al. Object-related activity revealed by
will be remembered. Science, 281(5380): 1185e1187, 1998. functional magnetic resonance imaging in human occipital
Buckner RL and Carroll DC. Self-projection and the brain. Trends in cortex. Proceedings of the National Academy of Sciences U.S.A.,
Cognitive Sciences, 11(2): 49e57, 2007. 92(18): 8135e8139, 1995.
Buckner RL, Andrews-Hanna JR, and Schacter DL. The brain’s McElroy L. The generation effect with homographs: Evidence for
default network: Anatomy, function, and relevance to disease. postgeneration processing. Memory and Cognition, 15(2):
Annals of the New York Academy of Sciences, 1124(1): 1e38, 2008. 148e153, 1987.
Cabeza R. Role of parietal regions in episodic memory retrieval: Metcalfe J and Kornell H. Principles of cognitive science in
The dual attentional processes hypothesis. Neuropsychologia, education: The effects of generation, errors, and feedback.
46(7): 1813e1827, 2008. Psychonomics Bulletin and Review, 14(2): 225e229, 2007.
Cabeza R and Nyberg L. Imaging cognition II: An empirical review Miller EK and Cohen JD. An integrative theory of prefrontal cortex
of 275 PET and fMRI studies. Journal of Cognitive Neuroscience, function. Annual Review of Neuroscience, 24: 167, 2001.
12(1): 1e47, 2000. Moss J, Schunn CD, Schneider W, McNamara DS, and VanLehn K.
Cansino S, Maquet P, Dolan RJ, and Rugg MD. Brain activity The neural correlates of strategic reading comprehension:
underlying encoding and retrieval of source memory. Cerebral Cognitive control and discourse comprehension. NeuroImage,
Cortex, 12(10): 1047e1056, 2002. 58(2): 675e686, 2011.
Cohen JD, Perstein WM, Braver BS, Nystrom LE, Noll DC, Jonides J, Paller KA and Wagner AD. Observing the transformation of
et al. Temporal dynamics of brain activation during a working experience into memory. Trends in Cognitive Sciences, 6(2):
memory task. Nature, 386(6625): 604e660, 1997. 93e102, 2002.
Craik FIM and Lockhart RS. Levels of processing: A framework for Poldrack RA, Wagner AD, Prull MW, Desmond JE, Glover GH, and
memory research. Journal of Verbal Learning and Verbal Behavior, Gabrieli JDE. Functional specialization for semantic and
11(6): 671e684, 1972. phonological processing in the left inferior prefrontal cortex.
D’Esposito M, Detrea JA, Aguirre GK, Stallcup M, Alsop DC, NeuroImage, 10(1): 15e35, 1999.
Tippet LJ, et al. A functional MRI study of mental image Postle BR. Working memory as an emergent property of the mind
generation. Neuropsychologia, 35(5): 725e730, 1997. and brain. Neuroscience, 139(1): 23e38, 2006.
Davachi L. Item, context and relational episodic encoding in Raichle ME, Macledo AM, Snyder AZ, Powers WJ, Gusnard DA, and
humans. Current Opinion in Neurobiology, 16(16): 693e700, 2006. Shulman GL. A default mode of brain function. Proceedings of
D’Esposito M, Postle BR, Ballard D, and Lease J. Maintenance the National Academy of Sciences U.S.A., 98(2): 676e682, 2001.
versus manipulation of information held in working memory: Raye CL, Johnson MJ, Mitchell KJ, Reeder JA, and Greene EJ.
An event related fMRI study. Brain and Cognition, 41(1): 66e86, Neuroimaging a single thought: Dorsolateral PFC activity
1999. associated with refreshing just-activated information.
Francis WN and Kuc era H. Frequency Analysis of English Usage: NeuroImage, 15(2): 447e453, 2002.
Lexicon and Grammar. Boston, MA: Houghton-Mifflin, 1982. Roediger HL and Karpicke HD. Test-enhanced learning: Taking
Hunt RR and McDaniel MA. The enigma of organization and memory tests improves long-term retention. Psychological
distinctiveness. Journal of Memory and Language, 32(4): 421e445, Science, 17(3): 249e255, 2006.
1993. Shimamura AP. The role of the prefrontal cortex in dynamic
Jacoby LL. Remembering the data: Analyzing interactive filtering. Psychobiology, 28(2): 207e218, 2000.
processes in reading. Journal of Verbal Learning and Verbal Shimamura AP. A neurocognitive approach to metacognitive
Behavior, 22(5): 485e508, 1983. monitoring and control. In Dunlosky J and Bjork R (Eds),
Johnson MK, Raye CL, Mitchell KJ, Greene EJ, Cunningham WA, Handbook of Memory and Metacognition. Mahwah, NJ: Psychology
and Sanislow CA. Using fMRI to investigate a component Press, 2008: 373e390.
process of reflection: Prefrontal correlates of refreshing a just- Shimamura AP. Hierarchical relational binding in the medial
activated representation. Cognitive, Affective, and Behavioral temporal lobe: The strong get stronger. Hippocampus, 20(11):
Neuroscience, 5(3): 339e346, 2005. 1206e1216, 2010.
Jurica PJ and Shimamur AP. Monitoring item and source Shimamura AP. Episodic retrieval and the cortical binding of
information: Evidence for a negative effect in source memory. relational activity. Cognitive, Affective, and Behavioral
Memory and Cognition, 27(4): 648e656, 1999. Neuroscience, 11(3): 277e291, 2011.
 c o r t e x 4 9 ( 2 0 1 3 ) 1 9 0 1 e1 9 0 9 1909

Simons JS, Davis SW, Gilbert SJ, Frith CD, and Burgess PW. Taconnat L, Baudouin A, Fay S, Clarys D, Vanneste S, Tournelle L,
Discriminating imagined from perceived information engages et al. Aging and implementation of encoding strategies in the
brain areas implicated in schizophrenia. NeuroImage, 32(2): generation of rhymes: The role of executive functions.
696e703, 2006. Neuropsychology, 20(6): 658e665, 2006.
Simons JS, Henson RN, Gilbert SJ, and Fletcher PC. Separable Thompson-Schill SL, D’Esposito M, Aguirre GK, and Farah MJ. Role
forms of reality monitoring supported by anterior prefrontal of left inferior prefrontal cortex in retrieval of semantic
cortex. Journal of Cognitive Neuroscience, 20(3): 447e457, 2008. knowledge: A reevaluation. Proceedings of the National Academy
Slamecka NJ and Graf P. The generation effect: Delineation of a of Sciences U.S.A., 94(26): 14792e14797, 1997.
phenomenon. Journal of Experimental Psychology: Human Tyler SW, Hertel PT, McCallum MC, and Ellis HC. Cognitive effort
Learning and Memory, 4(6): 592e604, 1978. and memory. Journal of Experimental Psychology: Human Learning
Smith RW and Healy AF. The time-course of the generation effect. and Memory, 5(6): 607e617, 1979.
Memory and Cognition, 26(1): 135e142, 1998. Uncapher MR and Wagner AD. Posterior parietal cortex and
Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, episodic encoding: Insights from fMRI subsequent memory
Behrens TEJ, Johansen-Berg H, et al. Advances in functional effects and dual-attention theory. Neurobiology of Learning and
and structural MR image analysis and implementation as FSL. Memory, 91(2): 139e154, 2009.
NeuroImage, 23(S1): S208eS219, 2004. van Veen V, Cohen JD, Botvinick MM, Stenger VA, and Carter CS.
Souliez L, Pasquier F, Lebert F, Leconte P, and Petit H. Generation Anterior cingulate cortex, conflict monitoring, and levels of
effect in short term verbal and visuospatial memory: processing. NeuroImage, 14(6): 1302e1308, 2001.
Comparisons between dementia of the Alzheimer type and Vilberg KL and Rugg MD. Memory retrieval and the parietal
dementia of the frontal lobe type. Cortex, 32(2): 347e356, 1996. cortex: A review of evidence from a dual-process perspective.
Spreng RN, Mar RA, and Kim SN. The common neural basis of Neuropsychologia, 46(7): 1787e1789, 2008.
autobiographical memory, prospection, navigation, theory of Wagner AD, Schacter DL, Rotte M, Koutstaal W, Maril A, Dale AM,
mind, and the default mode: A quantitative meta-analysis. et al. Building memories: Remembering and forgetting of
Journal of Cognitive Neuroscience, 21(3): 489e510, 2009. verbal experiences as predicted by brain activity. Science,
Taconnat L and Isingrini M. Cognitive operations in the 281(5380): 1188e1191, 1998.
generation effect on a recall test: Role of aging on divided Wilson MD. The MRC psycholinguistic database: Machine-usable
attention. Journal of Experimental Psychology: Human Learning dictionary, version 2.00. Behavioral Research Methods,
and Memory, 30(4): 827e837, 2004. Instruments, and Computers, 20(1): 6e11, 1988.