Sunteți pe pagina 1din 23

Journal of Experimental Psychology: General

© 2018 American Psychological Association 2019, Vol. 148, No. 8, 1312–1334


0096-3445/19/$12.00 http://dx.doi.org/10.1037/xge0000526

False Memory at Short and Long Term


Marlène Abadie and Valérie Camos
University of Fribourg

False memories are well-established long-term memory (LTM) phenomena. Recent reports of false
recognition at short term suggest that working memory (WM) could also give rise to false memories,
supporting the unitary view of memory. Alternatively, we hypothesized that the emergence of false
memories at short term results from the impairment of WM maintenance, memory performance relying
then on LTM. More specifically, we assumed that false memories rely on the retrieval of gist traces of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

the memory items while their verbatim traces that could block false memories are no longer accessible.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

To test this proposal, we reported a series of 4 experiments in which the availability of 2 WM


maintenance mechanisms, articulatory rehearsal and attentional refreshing, was manipulated, and the
reliance of recognition performance on gist and verbatim traces was also assessed. In line with our
hypothesis, the occurrence of false memories in immediate recognition test was accompanied by the
reduction of verbatim memory retrieval resulting from the impairment of rehearsal. By contrast, false
memories in the delayed test depended on gist memory, which was strengthened by the use of refreshing.
These findings support an integrated account of false memories at short and long term, shed light on the
nature of mental representations generated by WM maintenance mechanisms and on the relationships
between WM and LTM.

Keywords: false memory, working memory, long-term memory

Supplemental materials: http://dx.doi.org/10.1037/xge0000526.supp

The relationship between working memory (WM) and long- system for maintaining relevant representations across a relatively
term memory (LTM) is a central question for understanding hu- short interval, and LTM a virtually unlimited and permanent
man cognition and has been a matter of intense debate since the memory for events and general knowledge (e.g., Atkinson &
19th century. On the one hand, some models assume that WM and Shiffrin, 1968; Baddeley, 1986; Barrouillet & Camos, 2015). On
LTM are two distinct systems, with WM being a limited-capacity the other hand, models pertaining to the unitary-memory view
deny any distinction between WM and LTM (e.g., Brown, Neath,
& Chater, 2007; Nairne, 1988, 1990). Between these two extreme
This article was published Online First December 13, 2018.
theoretical positions, several intermediate models have been pro-
Marlène Abadie and Valérie Camos, Department of Psychology, Fri- posed that treat WM as a subset of LTM representations in a
bourg Center for Cognition, University of Fribourg. temporary state of heightened accessibility (e.g., Cowan, 1995;
This article is a collaboration between the two authors during the Oberauer, 2002).
postdoc of Marlène Abadie with Valérie Camos. Valérie Camos is an The present article will address the issue of the relationships
expert on working memory and she developed the Time-Based Resource- between WM and LTM by examining the impact of WM on the
Sharing model with Pierre Barrouillet, and especially demonstrated the occurrence of the false memory effect, a phenomenon that is
existence of distinct maintenance mechanisms. Marlène Abadie is an supposed to reflect the influence of LTM. Research on false
expert on long-term memory; she worked in particular on the distinction
memories has mainly focused on LTM, exploring how recall and
between verbatim and gist representations of the fuzzy-trace theory. This
study initiated a new line of research on the relationships between WM and
recognition of previously studied items can be distorted over a
LTM. Marlène Abadie is now at Laboratoire de Psychologie Cognitive, period ranging from minutes to days (Reyna, Corbin, Weldon, &
Aix-Marseille Université, France. Data of Experiment 1 and 2 were pre- Brainerd, 2016). In the Deese-Roediger-McDermott (DRM) task
sented at the 58th Annual Meeting of the Psychonomic Society and at the (Deese, 1959; Roediger & McDermott, 1995), participants are
20th European Society for Cognitive Psychology Conference. Part of this presented with lists including the 15 strongest word associates
work was supported by a grant to Valérie Camos from the Pool de (e.g., “sing, concert, piano, etc.”) for a given theme word (in this
Recherche of the Université de Fribourg. We thank Kim Hirsbrunner for example, “music”) and are then given a recall or recognition test.
her assistance with data collection. We also thank Clément Belletier, and Investigations using this paradigm consistently show that the un-
Laurent Waroquier for their advices, and Pierre Barrouillet for his advice
presented theme word is falsely remembered at very high levels,
and editing work.
Correspondence concerning this article should be addressed to Marlène
almost rivaling the acceptance rate of presented words (see Brain-
Abadie, who is now at the Laboratoire de Psychologie Cognitive, Centre erd & Reyna, 2005; Gallo, 2006, for reviews). Although the false
National de la Recherche Scientifique & Aix Marseille Université, memory phenomenon has been investigated largely in the LTM
UMR7290, Bâtiment 9 Case D, 3, Place Victor Hugo, 13331 Marseille domain, recent research also reported false recognition at short
Cedex 3, France. E-mail: marlene.abadie@univ-amu.fr term, suggesting that WM could also give rise to false memories
1312
FALSE MEMORY AT SHORT AND LONG TERM 1313

(e.g., Atkins & Reuter-Lorenz, 2008; Flegal, Atkins, & Reuter- particular the role of WM in preventing or enabling LTM influ-
Lorenz, 2010; Flegal & Reuter-Lorenz, 2014). According to At- ences. It also allowed to determine the nature of mental represen-
kins and Reuter-Lorenz (2008), the occurrence of semantic distor- tations that underlie these false memories, and more generally it
tions over the short and long term might indicate overlap between indicated how WM mechanisms give rise to different kinds of
WM and LTM mechanisms, thereby challenging any dissociation representations. In the following section, we present first the
between WM and LTM. theory that provided the most prominent account of long-term false
However, one might wonder whether showing that memory memories. Then, we review studies reporting false memories at
distortions can occur in a short-term test necessarily implies that short term. Next, we briefly describe WM maintenance mecha-
these distortions are specifically produced within WM. Indeed, nisms, their impact on short- and long-term tests, and the nature of
LTM can also influence short-term recognition, especially when representations they maintain. Finally, we propose and test in the
items cannot be maintained in WM. According to the primary- present series of experiments a new account of the role of WM
secondary memory framework (Unsworth & Engle, 2007), when maintenance mechanisms on the occurrence of false memories at
the number of items to be maintained in WM exceeds its capacity short and long delays.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

or when maintenance in WM is impeded, items are no longer


This document is copyrighted by the American Psychological Association or one of its allied publishers.

accessible in WM and must be retrieved from LTM for further


False Memory at Long Term
recall, even at short term. Supporting this view, it has been shown
that the level-of-processing (LOP) effect, a well-known LTM Several studies have shown that memory errors obtained using
effect (Craik & Lockhart, 1972; Craik & Tulving, 1975), appears the DRM paradigm are difficult to restrain (e.g., Gallo, Roberts, &
in a short-term recall test when WM maintenance is impaired and Seamon, 1997; McDermott, 1996). The fuzzy-trace theory (FTT;
that items are retrieved from LTM (e.g., Rose, Buchsbaum, & Brainerd & Reyna, 2002, 2005) provides one of the most promi-
Craik, 2014; Rose, Craik, & Buchsbaum, 2015). Hence, the mere nent account of the occurrence of long-term false memories (see
observation of false recognition in a short-term test is not sufficient also Gallo & Roediger, 2002; Robinson & Roediger, 1997; Roe-
to deny the existence of separate memory systems and to integrate diger, McDermott, & Robinson, 1998). According to the FTT, the
short and long-term memory processes within a single system. nature of representations retrieved during the test determines the
Providing compelling evidence that WM produces memory distor- formation of false memories. This model proposes that the surface
tion requires in a first place to ascertain that items are maintained form and meaning content of experiences are processed and stored
in, and retrieved from, WM. Otherwise, one cannot know whether in parallel. Verbatim traces are instantiated representations of the
short-term false memories come from WM influence as suggested surface forms of experienced items, whereas gist traces are inter-
by Atkins and Reuter-Lorenz (2008; Flegal et al., 2010; Flegal & pretations of concepts (meaning, patterns, relations) that are re-
Reuter-Lorenz, 2014). LTM influence, or mutual influence of both trieved as a result of encoding the surface form of the items. Both
memory systems. gist and verbatim retrieval support true memory for experienced
Our aim in the present study was thus to examine whether WM items, either because the corresponding items are specifically
mechanisms underlie the emergence of false memories at short and recollected (verbatim retrieval) or because their meaning is famil-
long delays. This question is of fundamental importance not only iar (gist retrieval). However, verbatim and gist retrieval have
to better understand the emergence of false memories, but also to opposite effects on false memories for items that preserve the
address the key issue of the structure and functioning of human meaning of experience (e.g., related distractors). Retrieving ver-
memory, which spans the entire field of cognitive psychology. For batim memories produces rejection of related distractors (remem-
this purpose, we manipulated the availability of mechanisms of bering “sing” or “concert” produces rejection of the unpresented
maintenance of information in WM and tested under which con- theme word “music”), whereas retrieving gist memories produces
ditions lists of associatively related words produce false memories acceptance of related distractors (remembering that the gist of the
in immediate and delayed recognition tests. We reasoned that if list was music produces acceptance of “music”).
false recognition errors appear at short term when maintenance of According to FTT, variations in the contributions of gist and
memory words in WM is disrupted, but disappear when mainte- verbatim memories determine responses to related distractors.
nance in WM is effective, then false memories are intrinsically Related distractors are good cues for gist memories and also better
under LTM influence and prevented by WM maintenance. By cues than unrelated distractors for verbatim memories (e.g., putting
contrast, WM would be the source of false memories if recognition “music” at test can remind people of “sing” from the study list, but
errors occur when WM mechanisms of maintenance are available, putting “apple” does not help reminding people of music-related
but disappear when these mechanisms are blocked. Finally, if false study items). Related distractors can cue the retrieval of verbatim
memories occur whatever the availability of WM maintenance memories, and retrieving verbatim memories produces recollection
mechanisms, then it might concluded that WM does not influence rejection of related distractors (Brainerd, Reyna, & Kneer, 1995).
false memories. According to FTT, access to verbatim and gist memories is as-
Two first experiments were conducted demonstrating the role of sumed to decline over time, but verbatim traces are faster to fade
WM in the occurrence of false memories: active maintenance of away. Therefore, when the delay between study and test is short,
information in WM prevented false memories in immediate test, verbatim traces are accessible and can oppose gist traces, reducing
but elicit their occurrence in delayed test. Two follow-up experi- the incidence of false memories. By contrast, at longer delays, the
ments were then conducted to disentangle the role of the different short-lasting verbatim traces should leave the more durable gist
WM maintenance mechanisms on the emergence of false memo- traces unopposed, and the incidence of false memories should
ries at short and long term. These two sets of experiments provided increase. Research on FTT also demonstrated that verbatim traces
insight on the relationships between WM and LTM, stressing in are more fragile and sensitive to cognitive load and interference
1314 ABADIE AND CAMOS

than gist traces (e.g., Reyna & Brainerd, 1995, for a review). immediate test and a surprise delayed recognition task that oc-
Furthermore, studies showed that performing a highly demanding curred after the presentation of all the lists. Taken together, these
concurrent task that generates an additional cognitive load during findings evidence that false memories can occur over the short
the retention interval between study and test has a higher detri- term. The existence of similar phenomena over the short and long
mental effect on the long-term retrieval of verbatim than gist term has been taken as strong evidence in favor of the unitary view
memory (Abadie, Waroquier, & Terrier, 2013, 2017). Therefore, of memory that does not distinguish between WM and LTM
there should be more false memories in delayed test when the (Flegal et al., 2010).
retention interval is filled with a highly attention-demanding rather However, it might be noted that the short-term false recognition
than a less demanding secondary task, because the verbatim traces effects reported in these studies appeared only when the number of
of the memory words would be less accessible to oppose gist memory items exceeded WM capacities (as in Coane et al., 2007)
traces. Finally, research on FTT showed that gist memory in- or when the retention interval between the study and test phases
creases when a concept has been repeatedly activated at study by was filled with a demanding distractor task that could have im-
multiple items as in the DRM lists (Brainerd, Reyna, & Mojardin, paired the maintenance of information in WM (Atkins & Reuter-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1999; Stahl & Klauer, 2008). Thus, false memories are more likely Lorenz, 2008; Flegal et al., 2010; Flegal & Reuter-Lorenz, 2014).
This document is copyrighted by the American Psychological Association or one of its allied publishers.

to occur when lists of associatively related words, rather than lists Hence, false memories seem to appear at short delays, but only
of unrelated words, are presented at study. when WM maintenance is not possible or particularly difficult.
In sum, according to the FTT, the occurrence of false memories Thus, it remains possible that WM maintenance mechanisms might
in delayed test depends on the nature of representations retrieved prevent short-term false memories instead of promoting them.
during the test, gist traces being the key ingredient in long-term
false memories. Moreover, the FTT predicts that false memories
Maintenance of Verbal Information in WM
are more likely to be observed when a concept has been repeatedly
activated at study by multiple items or when the retention interval Several mechanisms for the maintenance of information in WM
was filled with high rather than low demanding secondary task or have been described. In the time-based resource sharing (TBRS)
in delayed test rather than in immediate test. model of WM, two main mechanisms have been described as
responsible for the temporary maintenance of verbal information
in WM: articulatory rehearsal and attentional refreshing (see
False Memory at Short-Term
Camos, 2015, 2017, for reviews). They are assumed to be quali-
Although false memory effect has been largely investigated in tatively distinct because rehearsal is thought to operate by phono-
LTM, recent studies indicated that WM is also susceptible to logical repetition of memoranda, while refreshing is considered as
memory distortions by showing that false memories occur on the a domain-general, attention-based mechanism that operates by
timescale of a WM task (Atkins & Reuter-Lorenz, 2008; Coane, thinking back briefly to recently active memoranda (see Barrouil-
McBride, Raulerson, & Jordan, 2007; Flegal et al., 2010; Flegal & let & Camos, 2015; Camos, 2015, 2017; Camos & Barrouillet,
Reuter-Lorenz, 2014; Macé & Caza, 2011). Using a simple span 2014; Camos et al., 2018, for reviews).
task with DRM lists of three, five, or seven words, Coane, Importantly, the two mechanisms are thought to be independent
McBride, Raulerson, and Jordan (2007) asked participants to in- from each other. Camos, Lagner, and Barrouillet (2009) demon-
dicate whether a probe had been presented in the memory set or strated that manipulating the opportunity for rehearsal and refresh-
not. Probes were targets that had appeared in the memory set, ing had independent and additive effects on immediate recall. In a
related distractors that were associated with words in the memory complex span task, they manipulated refreshing by varying the
set (e.g., the theme word) or unrelated distractors that had not cognitive load (i.e., the attentional demand) of the secondary task,
appeared and were not associated with the memory set words. and rehearsal by requiring or not a concurrent articulation (i.e.,
Results showed that related distractors were more often falsely articulatory suppression). Manipulating the availability of one
recognized than unrelated distractors, although virtually no false mechanism while controlling for the other led to poorer immediate
recognition of related distractors occurred at Set Size 3. recall performance. Moreover, the orthogonal manipulation of
Similarly, Atkins and Reuter-Lorenz (2008) reported false mem- refreshing and rehearsal yielded additive effects, suggesting that
ories in a short-term Brown-Peterson recognition and a free-recall both mechanisms contribute to recall performance in an immediate
task (Brown, 1958; Peterson & Peterson, 1959) in which DRM test, but do so independently.
lists containing four associates were followed by a 3 to 4 s Further studies examined the role of each maintenance mecha-
retention interval that was filled with a mathematical distractor nism on LTM, providing evidence that the two mechanisms also
task. In the recognition task, the retention interval was followed by differ on their impact on long-term retention (Camos & Portrat,
one of the three types of probes (targets, related distractors, or 2015; Loaiza & McCabe, 2012; Rose et al., 2014). While manip-
unrelated distractors). Results showed that related distractors were ulations that affected refreshing reduced both immediate and de-
more often falsely recognized as members of the memory set than layed recall, the addition of a concurrent articulation reduced
unrelated distractors. Moreover, in the recall task, semantic errors immediate but not delayed recall. These findings indicate that both
(false recall of an unpresented theme word or an associated word) short- and long-term memory traces depend on the availability of
exceeded other errors. Although false recognition and false recall refreshing, whereas the use of rehearsal affects only short-term
still occurred, they were strongly reduced in the absence of verbal recall performance.
distraction during the retention interval. In a subsequent study, Given this distinction between rehearsal and refreshing in im-
Flegal, Atkins, and Reuter-Lorenz (2010) showed that false rec- mediate and delayed recall, other studies have investigated
ognition of related distractors occurred at the same rate in an whether the qualitative nature of representations differs as a func-
FALSE MEMORY AT SHORT AND LONG TERM 1315

tion of using each maintenance mechanism. Camos, Mora, and disrupts maintenance of verbatim traces of memory words but
Oberauer (2011; see also Camos, Mora, & Barrouillet, 2013; Mora leaves gist traces unaffected (Abadie et al., 2013, 2017). Moreover,
& Camos, 2013) showed that using rehearsal emphasized the Brainerd, Reyna, and Mojardin (1999) also showed that verbatim
phonological characteristics of the memory items. For example, memory decreases when memory words could not be repeatedly
the phonological similarity effect (i.e., poorer recall for phonolog- read and rehearsed. Thus, verbatim traces might be less accessible
ically similar words) and the word length effect (i.e., poorer recall when WM maintenance mechanisms are impeded, which should
for long words) appeared when rehearsal was used for maintaining increase false recognition of related distractors since the retrieval
memory words in a complex span task, but not when it was of gist traces is unopposed. So far, no studies have been conducted
suppressed (Camos et al., 2011; Mora & Camos, 2013). Therefore, to examine the role of WM maintenance mechanisms on false
rehearsal reinforces the phonological characteristics of the mem- memories. Thus, based on the FTT and TBRS model, we draw
ory words. Conversely, it was argued that refreshing operates some specific predictions regarding the role of each WM mainte-
independently of the phonological characteristics, but results are nance mechanism on the occurrence of short- and long-term false
mixed regarding the nature of representations that refreshing pro- memories.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

motes. Some studies suggested that semantic processing could be First, because rehearsal reinforces the phonological characteris-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

moderated by attention-based factors (Shivde & Anderson, 2011). tics of the memory words (e.g., Mora & Camos, 2013), has only
Another study showed a benefit of deep semantic processing transitory and no long-term effects (e.g., Camos & Portrat, 2015),
during encoding only when maintenance mechanisms of WM were and that the mere repetition of studied words fosters encoding and
prevented in an incidental encoding condition (Rose & Craik, retrieval of their surface forms independently of their meaning
2012). On the contrary, other studies provided evidence of seman- (Loaiza & Camos, 2018), rehearsal should strengthen verbatim
tic processing effects in immediate recall independently of the traces in WM. Indeed, verbatim memories are representations of
availability of refreshing. Loaiza and Camos (2018) presented a the surface forms (e.g., phonological characteristics) of memory
cue word that was semantically or phonologically related to a items, which cannot be maintained over the long term (e.g., Reyna
target during the recall phase of a complex span task. Results & Brainerd, 1995). Furthermore, rehearsal emphasizes the phono-
showed a benefit of semantic over phonological retrieval cues, but logical, but not the semantic, characteristics of the memory words
this benefit did not vary with the cognitive load of the concurrent (Loaiza & Camos, 2018; Rose et al., 2014, 2015). Hence, rehearsal
task or the intention to learn (see also Craik & Tulving, 1975; should not allow maintenance of gist memories, which represent
Hyde & Jenkins, 1973). the meaning of information. In addition, gist traces persist over
In sum, two mechanisms allow the maintenance of verbal in- time whereas rehearsal cannot create long-lasting memories (e.g.,
formation in WM. Rehearsal can be prevented by concurrent Camos & Portrat, 2015). Thus, there should be no false recognition
articulation, which reduces immediate recall but not delayed recall of related distractors in an immediate test when participants use
performance. Refreshing is impaired by increasing the cognitive rehearsal to maintain memory words. By contrast, false memories
load of a concurrent task, resulting in reduced immediate and should appear when rehearsal is prevented. Findings of a unique
delayed recall performance. Finally, both mechanisms differ in the experiment that investigated the effect of articulatory suppression
nature of the representations they promote. Rehearsal reinforces on short-term false memories are congruent with this hypothesis.
the phonological characteristics of the memory items, whereas it is Macé and Caza (2011) used a short-term recognition task in which
less clear what are the representations that refreshing can manip- participants had to judge whether two successive mixed lists
ulate. consisting of related and unrelated words were matched. They
found that, relative to silence, the addition of a concurrent articu-
A Theoretical Account of the Role of WM lation during lists presentation increased false recognition of mis-
matching lists comprising a critical lure (e.g., a word that was
Mechanisms on False Memory
semantically associated with all the words of the first list). Finally,
By integrating the FTT and the TBRS model, we propose a new regarding the role of rehearsal on long-term false memories, as
theoretical account of the relationships between WM and LTM rehearsal does not impact the retention over the long term, it
that explains the emergence of short- and long-term false memo- should not impact delayed false recall and recognition. Whether
ries. Our reexamination of previous research (Atkins & Reuter- rehearsal was used or not, verbatim memories decline over time,
Lorenz, 2008; Flegal et al., 2010; Flegal & Reuter-Lorenz, 2014) and remembering after a long delay could not be primarily based
suggests that false memories occur in immediate test when the use on the retrieval of verbatim memory, but mostly on the retrieval of
of WM maintenance mechanisms is impeded. In studies in which gist memory or on guessing.
participants maintained word lists while verifying a math equation, Second, contrary to rehearsal, refreshing allows the maintenance
the use of rehearsal is likely to have been disrupted, because of information not only at short, but also at long delays (Camos &
equation verification is known to rely on the phonological loop Portrat, 2015). Although it remains unclear whether refreshing
(DeStephano & LeFevre, 2004; Fürst & Hitch, 2000). Moreover, manipulates semantic representations (Loaiza & Camos, 2018;
the equation verification task is attention-demanding and could Shivde & Anderson, 2011), it is clear that it does not only empha-
therefore reduce the ability to refresh the memory items (Ver- size verbatim representations since it impacts LTM remembering.
gauwe, Camos, & Barrouillet, 2014). Because preventing any WM Thus, one could assume that refreshing would strengthen both gist
maintenance mechanism impairs immediate recall performance, it and verbatim memories. FTT (Reyna, 2012) predicts that when
might also boost false recognition of related distractors. Indeed, both verbatim and gist representations are equally available, adults
studies on FTT indicated that completing an attention-demanding tend to rely more on gist representations to generate responses (i.e.,
secondary task during the retention interval between study and test a fuzzy-processing preference). If refreshing allows maintenance
1316 ABADIE AND CAMOS

of both the meaning and surface characteristics of memory words, Four experiments were conducted in which the attentional de-
and that people should rely primarily on the meaning to make mand of the concurrent task and the presence (or not) of a con-
recognition judgments, then false recognition of related distractors current articulation during the retention interval were manipulated
is likely to occur in immediate test when refreshing is available. By to impede attentional refreshing and articulatory rehearsal, respec-
contrast, preventing refreshing should reduce the retrieval of both tively. In the first experiment, both maintenance mechanisms
verbatim and gist memory. However, because verbatim memory is could be used; in the second experiment, both mechanisms were
more sensitive to cognitive load than gist memory (Abadie et al., impeded; in the third one, refreshing was available and rehearsal
2013, 2017), preventing refreshing should affect verbatim memory was blocked whereas in the fourth experiment, only rehearsal was
more than gist memory. Hence, there should be more false mem- available, refreshing being impaired. The two first experiments
ories in the immediate test when refreshing is prevented than when were designed to test the role of WM maintenance on false
it is available. Obviously, this would be the case only if memory memories and the two last to disentangle the effects of refreshing
words cannot be maintained using rehearsal, their verbatim traces and rehearsal.
being otherwise preserved by this mechanism. Moreover, as re- The second aim of these experiments was to examine the con-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

freshing reinforces gist memories during maintenance, these mem- tribution of verbatim and gist memory to the false memory effect
This document is copyrighted by the American Psychological Association or one of its allied publishers.

ory traces can be preserved over the long term. Thus, the use of at short and long delays. To gain insight into these memory
refreshing should foster the occurrence of false memories in de- representations, we used the simplified version of the conjoint-
layed test as well. Conversely, like true memories, the occurrence recognition (CR) model of the FTT (Brainerd et al., 1999; Stahl &
of false memories at long delays should be reduced when refresh- Klauer, 2008). The CR model is a mathematical model that uses
ing is prevented. performance obtained in a recognition task to extract direct mea-
In sum, using rehearsal to maintain memory words should surements of verbatim and gist representations (see Brainerd et al.,
prevent the occurrence of false memories in immediate tests, 1999, for a complete description of the CR model, and Stahl &
whereas using refreshing should foster their occurrence in delayed Klauer, 2008, for a complete description of the simplified version).
tests. This makes the relationships between WM maintenance The model provides parameter estimates for verbatim and gist
mechanisms and the emergence of false memories more complex memory for the responses (i.e., acceptance or rejection) to the
than it could have been expected, as WM mechanisms could either different types of probes. CR estimates verbatim and gist memory
prevent or promote false memories depending on which mecha- that underlie the true recognition of targets, the correct rejection of
nism is considered, and for which delay of retention. related distractors and the false recognition of related distractors,
respectively, and also provides guessing parameters computed
from acceptance and rejection rate of unrelated distractors.
The Present Study
The first experiment tested our main hypothesis that false mem-
Our present purpose was to test this functional account of the ory is not an inherent WM phenomenon and there should be no
emergence of false memories at short and long delays. Our meth- false recognition of related distractors in the immediate test if both
odology was similar to the one used in previous studies reporting refreshing and rehearsal could be used to maintain memory words
short-term false memories (Atkins & Reuter-Lorenz, 2008; Flegal in WM. Indeed, both mechanisms should enhance the retrieval
et al., 2010; Flegal & Reuter-Lorenz, 2014), namely a Brown- of the verbatim traces of targets, resulting in correct rejection of
Peterson task with both an immediate and a delayed recognition related distractors in the immediate test. False recognition of
test. We manipulated the associative similarity of the word lists, related distractors, however, should appear in the delayed test
which served as memory sets. Memory sets consisted of four because refreshing should also enhance the retrieval of gist mem-
associatively related words (high associative similarity condition) ory, which is more durable than verbatim memory. The second
or four unrelated words (low associative similarity condition). This experiment should support the key conclusions of the first one by
manipulation has been shown to affect gist memory (e.g., Brainerd showing that when WM maintenance is prevented, false recogni-
et al., 1999). It was implemented here to manipulate the probability tion appears in the immediate test. Indeed, preventing both WM
of emergence of false memories. The presentation of each word list maintenance mechanisms should reduce the retrieval of both ver-
was followed by a 4-s retention interval. Half of the word lists batim and gist traces, but verbatim traces should be more affected
were probed in the immediate recognition test and the other half than gist traces, leading to some false recognition. Moreover, there
were probed in the delayed test. This procedure allows to compare should be few or no false recognition of related distractors in the
performance across equivalent lists without confounding long- delayed test, because both verbatim and gist traces should be so
term recognition with intervening short-term recognition of the weak that they could not be preserved over the long term.
same list (adapted from Flegal et al., 2010). There were three probe The two last experiments examined the role of each mainte-
types: target, unrelated distractor, and related distractor. To obtain nance mechanism separately. In Experiment 3, only rehearsal was
comparable measures of memory performance across delays, we prevented to investigate the role of refreshing on false memories.
computed an index for false recognition by subtracting unrelated Because refreshing should enhance the retrieval of both gist and
distractors from related distractors error rates (producing the dis- verbatim memory, we expected false recognition of related dis-
criminability index) as done by Flegal et al. (2010). Then, we tractors in both immediate and delayed tests when refreshing is
compared high associatively similar word lists to low associatively available. Finally, only refreshing was prevented in Experiment 4
similar word lists on this index to assess false memories. Larger to examine the role of rehearsal on false memories. We assumed
values of this index in the high associative similarity condition that rehearsal would enhance the retrieval of the verbatim traces
relative to the low associative similarity condition indicated the of targets in the immediate test, which would prevent false recog-
presence of false memories. nition of related distractors. Rehearsal should not affect recogni-
FALSE MEMORY AT SHORT AND LONG TERM 1317

tion in the delayed test. However, there should be no false recog- course credits or a cinema ticket. Participants were randomly
nition in the delayed test because refreshing is not available to assigned to the high or the low associative similarity condition.
create strong gist traces, and also because rehearsal should rein- One participant was mistakenly assigned to the low associative
force verbatim memory at the expense of gist memory. similarity condition resulting in an unbalanced number of partic-
For all experiments, we also reported true recognition scores. As ipants between conditions. All participants performed the imme-
previously observed, we expected that both refreshing and re- diate and the delayed recognition tasks.
hearsal increase true recognition in the immediate test, whereas Materials.
only refreshing would increase true recognition in the delayed test. Item recognition task. Two types of lists were constructed:
We also examined the nature of representations that underlie true high associatively similar and low associatively similar lists. To
recognition. We expected that rehearsal would strengthen verbatim create the lists, 64 theme words were selected from the verbal
memory for targets whereas refreshing would strengthen both gist association norms for concrete French nouns (Bonin, Méot, Fer-
and verbatim traces for targets. rand, & Bugaïska, 2013). Each theme word was (a) a high-
frequency and concrete noun, (b) had at least four associates, and
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Experiment 1 (c) shared no associates with any other theme words. Sixteen of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

these theme words were chosen to be the unrelated distractors and


In Experiment 1, participants maintained word lists in a Brown- were never presented as memory items. For each of the remaining
Peterson task, while they concurrently performed a low attention- 48 theme words, a high associatively similar list was then con-
demanding secondary task that did not elicit any concurrent artic- structed by taking the first four most highly associated words (e.g.,
ulation, thus preserving the use both rehearsal and refreshing. rooster, chick, hen, wing) of these theme words (e.g., chicken).
Regarding false recognition, we expected that both rehearsal and The low associatively similar lists were created by replacing the
refreshing would stimulate the retrieval of verbatim traces of the second, third, and fourth words of each of the high associatively
targets, which would oppose gist memory and minimize false similar lists (e.g., chick, hen, wing for the list chicken) by three
recognition of related distractors in the immediate test. Hence, new words, each of them being the most highly associated word to
false recognition rate should be small and should not differ as a 144 other theme words (e.g., rooster, whip, drawer, cave). In this
function of associative similarity of the word lists. Because ver- way, the four words in each low associatively similar list were
batim memory that underlies correct rejection of related distractors unrelated and each low associatively similar list had one word in
is assumed to decline over time, participants should rely more on common with the high associatively similar lists. Pilot testing
gist memory when a related distractor is presented in the delayed using the same procedure as Atkins and Reuter-Lorenz (2008)
test. Moreover, because the use of refreshing would enhance gist
confirmed that our high associatively similar lists, but not our low
memory, we expected to replicate the traditional false memory
associatively similar lists, induced false recognition of unstudied
effect in the delayed test with a larger rate of false recognition of
theme words in an immediate test. Table 1 provides a description
related distractors for lists of associatively related words (high
of both types of lists.
associative similarity condition) than for lists of unrelated words
Concerning the probes, there were three types of probes: related
(low associative similarity condition).
distractors, unrelated distractors, and targets. Each type of probes
Regarding true recognition, we predicted that it would only be
was presented in one third of the trials. In related distractor trials,
affected by the time of test. True recognition should be high in the
the probe was the theme word of the list in the high associative
immediate test because both rehearsal and refreshing could be used
similarity condition (e.g., chicken for the list rooster, chick, hen,
to maintain memory words. It should be reduced in the delayed test
wing), and a theme word related to one of the four presented words
because of time-related decay of verbatim and gist memory for
in the low associative similarity condition (e.g., chicken for the list
targets. We expected however that the effect of time of test would
rooster, whip, drawer, cave). In unrelated distractor trials, the
be smaller on gist memory, which is more durable than verbatim
probe was one of the 16 theme words selected from French
memory and could be strengthened by the use of refreshing.
Finally, we expected that gist memory for targets would be larger association norms and for which no lists were created. Finally, in
in the high than in the low associative similarity condition because target probe trials, we replicated Atkins and Reuter-Lorenz’s
multiple activations of the same concept is assumed to increase (2008, 2010) method. In the high associative similarity condition,
gist memory (e.g., Brainerd et al., 1999). the first associate of each list was replaced by the theme word for
that list, subsequently presented as probe (e.g., rooster was re-
placed by chicken). Similarly, in the low associative similarity
Method1 condition, one of the four words was replaced by its theme word
Sample size. We computed a meta-analytic effect size across presented at test as probe. In this way, in related distractors and
studies that demonstrated the occurrence of false memories within target probes trials, the probes were the same words for each list.
several seconds of studying brief four-item lists (Atkins et al., Per participant, a given probe appeared only once during the
2008; Flegal et al., 2010). The meta-analytic effect size was experiment.
Cohen’s d ⫽ 1.057, 95% CI [0.751, 1.362]. The procedure used in In each associative similarity condition, lists were divided into
these studies was similar to the present one. Power analysis indi- three groups of 16 lists each. Each group was associated to one
cated that 50 participants would be needed to achieve a 95% power
(GⴱPower; Faul, Erdfelder, Buchner, & Lang, 2009). 1
The four experiments reported in this article have been approved by the
Participants and design. Fifty participants (12 males, Mage ⫽ internal ethics committee of the department of psychology of the Univer-
20.7 years, SDage ⫽ 1.6) took part in Experiment 1 for either sity of Fribourg.
1318 ABADIE AND CAMOS

Table 1
Description of Word Lists and Probes Presented as a Function of the Associative Similarity Condition and Type of Probes in the
Recognition Test

High similarity condition Low similarity condition


Probe type Word list Probe Word list Probe

Related First four most highly associated Theme word The most highly associated word Theme word related to one
distractor words to a theme word e.g., “chicken” of four different theme words word
selected from Bonin et al.’s selected from Bonin et al.’s e.g., “chicken”
(2013) norms. norms.
e.g., “rooster, chick, hen, wing” e.g., “rooster, whip, drawer, cave”
Unrelated First four most highly associated Theme word selected The most highly associated word Theme word selected from
distractor words to a theme word from Bonin et al.’s of four different theme words Bonin et al.’s (2013)
selected from Bonin et al.’s norms and that selected from Bonin et al.’s norms and that was not
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(2013) norms. was not used in (2013) norms. used in our material
This document is copyrighted by the American Psychological Association or one of its allied publishers.

e.g., “rooster, chick, hen, wing” our material e.g., “rooster, whip, drawer, cave” e.g., “rain”
e.g., “rain”
Target A theme word with its first three Theme word A theme word plus the first Theme word related to one
most highly associated words e.g., “chicken” associated word from three other word
selected from Bonin et al.’s theme words selected from e.g., “chicken”
(2013) norms. Bonin et al.’s (2013) norms.
e.g., “chicken, chick, hen, wing” e.g., “chicken, whip, drawer, cave”

type of probes, the group-probe association being counterbalanced For the immediate test trials, a probe word appeared on screen
across participants. Each group had the same mean backward for 3,000 ms after the secondary task, and participants indicated
associative strength (M ⫽ 0.29) for the high associatively similar whether it was in the memory set by pressing the S (yes) or L (no)
lists. Memory lists and mean backward association strength for keys on a standard keyboard (Figure 1A). When they responded
each of them are available at https://osf.io/ets8j/. “no,” they had to indicate within the next 3,000 ms whether the
Secondary task. Stimuli were adapted from Vergauwe, Bar- probe was related to a studied word by pressing the same keys (yes
rouillet, and Camos’ (2009) spatial fit task. They consisted of a set or no). In trials that were not probed in the immediate test, a
of 24 white boxes, each of which contained a black horizontal line geometrical shape (a square, a circle or a diamond) replaced the
and two square black dots. The horizontal line was displayed in the probe word, and participants had 3,000 ms to indicate whether it
center of each box and the dots were positioned on the same was a square (yes or no; Figure 1B). When they responded “no,”
horizontal plane as each other, either above or below the horizontal a second geometrical shape appeared, and they had to indicate
line. The line varied in length, and the distance between the dots within 3,000 ms whether it was a square (yes or no). This proce-
was chosen so that for half of the boxes the line could fit into the dure was used to prompt the same type of responses and to keep
gap between the dots. The stimuli were presented one at a time. retention duration constant in all trials. Each participant completed
Procedure. To avoid confounding delayed recognition perfor- 48 trials presented in random order. In the 24 trials probed in the
mance with an intervening immediate recognition of the same lists, immediate test, eight were of each probe type (related distractor,
the procedure included two types of trials. For half of the trials, the unrelated distractor, target).
memory lists were probed within the same trial (i.e., immediate Following the completion of all trials, participants had to count
test) and for the remaining trials, they were probed in a recognition backward by twos from a given three-digit number for 2 min. This
test following the completion of all trials (i.e., delayed test).
distractor task was performed aloud to prevent any use of main-
At the beginning of the experiment, participants were informed
tenance strategies. Then, participants received instructions for the
that they would see several word lists and that they would com-
delayed recognition test. Each participant completed 24 trials,
plete two recognition tests, one immediately after the presentation
which tested memory lists that had not been probed in the imme-
of each list and one at the end of the experiment after the presen-
diate test (eight of each probe type). In each trial, a probe word
tation of the last list. Half of the participants were presented with
appeared for 3,000 ms, and participants indicated whether it was
high associatively similar word lists whereas the other half were
presented during the presentation of the lists or not by pressing
presented with low associatively similar word lists. Each trial
keys. When they responded “no,” they had 3,000 ms to indicate
began with the four words of a memory list, which were presented
simultaneously for 1,200 ms (see Figure 1). Participants were whether the probe was associated with a studied word or not by
asked to read the words aloud. Then, during a 4,000-ms retention pressing the same keys.
interval, participants completed a secondary task on each stimulus
that was displayed for 862 ms with interstimuli interval (ISI) of Results
431 ms. Participants were asked to press the space-bar as quickly
and as accurately as possible when a stimulus appeared. This task We verified that all participants complied with instructions and
was a simple reaction time (RT) task known for its low attention did press the space-bar at each occurrence of stimuli in the sec-
demand (Barrouillet, Bernardin, Portrat, Vergauwe, & Camos, ondary task. We conducted Bayes factor analyses, first to assess
2007) and performed without any concurrent articulation. the effect of time of test and associative similarity on memory
FALSE MEMORY AT SHORT AND LONG TERM 1319
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 1. Diagram of the experimental procedure. (A) Immediate test trials. (B) Delayed test trials.

accuracy, and then on gist and verbatim memory. Data of the four evidence in favor or against an effect is quantified by comparing a
experiments are available at https://osf.io/ets8j/. model including the effect (H1) to a model omitting it (H0). The
Memory accuracy. We computed discriminability indexes to relative likelihood of the two models under comparison is the
obtain comparable measures of memory performance across delay Bayes factor (BF; Jeffreys, 1961; Kass & Raftery, 1995). BF
(see Flegal et al., 2010). True and false recognition were condi- indicates the factor by which prior odds ratio for the two models
tionalized by subtracting the baseline false recognition rate of should be multiplied to obtain the posterior odds ratio. The odds
unrelated distractors (i.e., responses “target” or “related” to unre- ratio represents our degree of belief that the data was generated
lated distractors) from the rate of correct recognition of targets from one model (e.g., H1) relative to the other (H0). For instance,
(true recognition) and from the rate of false recognition of related if we think that both models are equally probable before seeing the
distractors as targets (false recognition). Larger values of the data (i.e., our prior odds ratio is 1) and BF in favor of H1 is 10, then
discriminability index of false recognition in the high relative to the posterior odd ratio is 10, which means that observed data are
the low associative similarity condition indicated the presence of 10 times more likely to have occurred under H1 than under H0.
false memories. True and false recognition data are shown in The BF can be used as a continuous index of the strength of
Figure 2. Response frequencies are given in the online supplemen- evidence for one model relative to another. The BF can express the
tary material. evidence in favor of H1 over H0 (BF10) or the evidence of H0 over
The true and false recognition data were submitted to default H1 (BF01). Here we reported BF10.
Bayesian repeated-measures ANOVA with time of test as For our Bayesian 2 (Time of Test) ⫻ 2 (Associative Similarity)
repeated-measures factor and associative similarity as between- ANOVA, five models are considered: (a) a null model in which
subjects factor (Rouder, Morey, Speckman, & Province, 2012). there was no effect of time of test nor of associative similarity, (b)
Bayesian analyses were conducted using JASP (Version 0.8.3.1, a model in which there was an effect of time of test but not of
JASP team, 2018). In Bayesian Hypothesis Testing, the strength of associative similarity, (c) a model in which there was an effect of
associative similarity but not of time of test, (d) an additive model
in which there was an effect of both variables but no interaction
True recognition False recognition
High similarity Low similarity between them, and (e) a full model in which there was an effect of
1
both variables and an interaction. The Bayes factors for each
0.8 model relative to the null can be found in Table 2. A BF10 of 3 or
more is considered substantial evidence for the model of interest,
0.6
a BF10 below one third is considered substantial evidence for the
0.4 null model and values around 1 indicate no substantial evidence
0.2
either way (Dienes, 2014; Jeffreys, 1961). In the text, first, we
reported the best model, that is, the model with the largest BF10.
0 Then, we compared the best model with each of the other models
Immediate test Delayed test Immediate test Delayed test
to check if it is substantially preferred to all of them. In case it was
Figure 2. True and false recognition accuracy as a function of time of test not substantially preferred, we reported also the second-best
and associative similarity in Experiment 1. Error bars represent standard model. When the full model is preferred, we conducted further
deviations. analyses to interpret the interaction.
1320 ABADIE AND CAMOS

Table 2
Summary of Bayesian Analyses (BF10) for the Four Experiments

Models
Experiments Measures Time of test Associative similarity Additive model Full model

Experiment 1 True recognition 9.83e ⴙ 17 .23 2.45e ⫹ 17 6.67e ⫹ 16


False recognition .25 53.3 13.7 4216
Vt 4.08e ⫹ 23 3.86 4.20e ⫹ 26 2.23e ⴙ 31
Gt 6.31e ⫹ 6 9.91e ⫹ 9 9.35e ⫹ 18 6.28e ⴙ 19
Vr 751 4.69 5608 3.44e ⴙ 7
Gr 1.94e ⫹ 19 6.29e ⫹ 7 1.87e ⫹ 40 1.91e ⴙ 48
Experiment 2 True recognition 7.52e ⴙ 9 .24 2.20e ⫹ 9 1.89e ⫹ 9
False recognition 3029 2.69 12092 91696
Vt 1.92e ⴙ 15 .25 5.14e ⫹ 14 3.41e ⫹ 14
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Gt 7926 291 9.03e ⫹ 6 9.60e ⴙ 13


2.33e ⫹ 14 1.43e ⫹ 15 5.83e ⴙ 16
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Vr 1.28
Gr 1.36e ⫹ 7 461 1.60e ⫹ 11 6.21e ⴙ 14
Experiment 3 True recognition 1.38e ⴙ 16 .27 4.58e ⫹ 15 1.29e ⫹ 15
False recognition 8.42 250 2737 1014
Vt 5.64e ⴙ 30 .31 4.33e ⫹ 30 2.11e ⫹ 30
Gt 719298 567 3.77e ⫹ 9 5.12e ⴙ 9
Vr 1.15 431 620 195
Gr 1.59e ⫹ 14 425793 1.06e ⴙ 25 3.92e ⫹ 24
Experiment 4 True recognition 3.09e ⫹ 19 .48 4.02e ⫹ 19 8.49e ⴙ 19
False recognition 2.17 .24 .51 .25
Vt 2.25e ⫹ 25 .43 2.61e ⴙ 25 7.86e ⫹ 24
Gt 2.66e ⫹ 21 6.51 1.74e ⫹ 25 1.51e ⴙ 43
Vr 1.02e ⫹ 18 .53 1.34e ⴙ 18 5.28e ⫹ 17
Gr 6.26e ⫹ 13 421 6.55e ⫹ 19 1.35e ⴙ 31
Note. Vt ⫽ verbatim memory for targets; Gt ⫽ gist memory for targets; Vr ⫽ verbatim memory for related distractors; Gr ⫽ gist memory for related
distractors. The BF10 for the null model was 1. BF10 for the best model is in bold.

True recognition. The model that only included time of test the target given the same cue. Vt and Gt underpin true recognition
was the best model, true recognition rate being higher in the of targets, reinforcing each other for true recognition. The param-
immediate than in the delayed test (Cohen’s dunb ⫽ 2.551, 95% CI eter Vr represents the probability of retrieving the verbatim trace of
[2.006, 3.169]2). a target when a related distractor is given as retrieval cue, whereas
False recognition. The full model was the best model. Be- Gr represents the probability of retrieving the gist trace a target
cause follow-up Bayesian t test showed a substantial preference for given the same cue. Vr and Gr underpin false recognition of related
a null effect of time of test, similar tests were performed to assess distractors, but they oppose each other in suppressing or fomenting
the effect of associative similarity separately in the immediate and false memories. The guessing parameter b assesses the tendency to
the delayed tests. As predicted, there was a substantial preference identify targets, related distractors and unrelated distractors as
for a null effect of associative similarity in the immediate test studied words. Parameter ␣ assesses the tendency to identify any
(BF10 ⫽ 0.3), and a substantial preference for an effect of asso- type of probe as a target rather than as a related distractor. The
ciative similarity in the delayed test, the rate of false recognition model equations are provided in the appendix of Stahl and Klau-
being higher in the high than in the low associative similarity er’s (2008) article.
condition (BF10 ⫽ 4060, Cohen’s dunb ⫽ 1.486, 95% CI [0.873, Consider the processing-tree representation of the model given
2.135]) in this test. These results replicated those of classic false in Figure 3. The first tree represents the processes occurring when
memory studies (e.g., Roediger & McDermott, 1995) showing a target is presented at test. When verbatim memory is available,
high rate of false recognition of related distractors in a delayed test, it is correctly identified as a target. When there is no verbatim
but contrasted sharply with more recent studies (e.g., Atkins &
memory but available gist memory, participants have identified the
Reuter-Lorenz, 2008; Flegal et al., 2010) showing the false mem-
probe meaning as old, but they cannot remember whether the
ory effect in both immediate and delayed tests.
probe itself or a related word with the same gist had been presented
Verbatim and gist memory. We used the simplified CR
in the study phase. A decision has to be made between the
model of the FTT (e.g., Stahl & Klauer, 2008) to compute param-
eter estimates for verbatim memory for targets (Vt), gist memory
for targets (Gt), verbatim memory for related distractors (Vr), and 2
Following recommendations by Cumming (2014) and Lakens (2013),
gist memory for related distractors (Gr) as well as parameters for we used the average standard deviation of both repeated measures and the
guessing processes (␣ and b) for each of the four experimental pooled standard deviation of both groups of measures as the standardizer
for calculating Cohen’s d for within-subject and between-subject measure-
conditions. The parameter Vt represents the probability of retriev- ments, respectively. Then, we apply Hedges’ correction to get an unbiased
ing a verbatim trace of a target given a target as retrieval cue, estimation Cohen’s dunb. The calculation of Cohen’s dunb and the CI on d
whereas Gt represents the probability of retrieving a gist trace of was carried out on ESCI (Cumming, 2014).
FALSE MEMORY AT SHORT AND LONG TERM 1321
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 3. Processing-tree for the simplified conjoint recognition paradigm based on Stahl and Klauer (2008).
Rectangles on the left represent probe types and rectangles on the right represent responses to questions “Was
the word in the memory set?” and “Was the word related with a word in the memory set?” They are connected
by branches of the processing tree that represent the combination of cognitive processes postulated by the model.
Vt ⫽ verbatim memory for targets; Gt ⫽ gist memory for targets; Vr ⫽ verbatim memory for related distractors;
Gr ⫽ gist memory for related distractors; b ⫽ guessing that an item is either a target or a related distractor; ␣ ⫽
guessing “target.”

responses “target” and “related.” The probe is identified as a target class weighted by the individual probabilities of belonging to each
with the probability a and as a related distractor with the proba- class. We checked they were similar to the parameter estimates in
bility 1 ⫺ a. When neither verbatim nor gist memory is available, each condition computed from the single-class model. For each
participants can still guess that the probe meaning is old with the experiment, the omnibus test of goodness of fit showed that the
probability b. A decision between the responses “target” and single class model fit was very good, and we reported the value of
“related” is required, which is again modeled by the parameter a. G2 in footnote for each experiment.3 Bayesian analyses were used
They also can guess that the probe is new with the probability 1 ⫺ to test the effect of time of test and associative similarity on the
b. The second tree represents the processes occurring when a model parameters. The results of Bayesian analyses are provided
related distractor is presented at test. Given available verbatim on Table 2. The analyses on verbatim and gist memory are re-
memory, it is correctly identified as a related distractor. Retrieving ported in the next subsection and those on guessing parameters are
gist memory leads participants to identify the probe meaning as old provided in the online supplementary material.
but would not allow them to remember whether the probe itself or Verbatim and gist memory for targets. Regarding Vt, the full
a related word with the same gist had been studied. With the model was the best model. Follow-up tests showed a substantial
probability a, the related distractor is identified erroneously as a preference for an effect of associative similarity in the delayed test
target and, with the probability 1 ⫺ a, it is identified as a related (BF10 ⫽ 70,998, Cohen’s dunb ⫽ 1.687, 95% CI [1.056, 2.359]), Vt
distractor. When no memory traces are available, participants can being larger in the low associative similarity condition, and a
guess that the probe meaning is old with the probability b or new marginal preference for a null effect in the immediate test (BF10 ⫽
with the probability 1 ⫺ b. Finally, as shown in the third tree, 0.35). Regarding Gt, the full model was also the best model.
identification of unrelated distractors is based on a combination of Follow-up analyses provided substantial evidence that Gt was
guessing processes a and b. larger in the high than in the low associative similarity condition in
Verbatim, gist, and guessing parameter estimates are given in both the immediate (BF10 ⫽ 340836, Cohen’s dunb ⫽ 1.896, 95%
Table 3. Individual parameter estimations were computed using CI [1.244, 2.594]) and the delayed test (BF10 ⫽ 1.004e ⫹ 12,
hierarchical modeling (Klauer, 2006), and latent-class estimates Cohen’s dunb ⫽ 3.089, 95% CI [2.294, 3.967]).
using Stahl and Klauer’s (2008) HMM tree. When individual Verbatim and gist memory for related distractors. Regarding
parameters differ (because some individuals having better verba- Vr, the full model was the best model. Substantial evidence was
tim memory than others), several latent classes can be estimated, gathered that Vr was larger in the high than in the low associative
and each individual is sorted into one of these classes (e.g., Person
similarity condition in the immediate test (BF10 ⫽ 615, Cohen’s
i belongs to Class C1 with the probability p, and to Class C2 with
the probability 1 ⫺ p). Before these analyses, we computed indi-
vidual parameter estimates as an average of the parameter of each 3 2
G(df ⫽12) ⫽ 1.49, p ⫽ 1.
1322 ABADIE AND CAMOS

Table 3
Parameter Estimates for Verbatim and Gist Memory and Guessing Processes as a Function of Time of Test and Associative Similarity
in Experiment 1

Immediate test Delayed test


Parameters High similarity Evidence for Low similarity High similarity Evidence for Low similarity

Vt .87 (.07) ⫽ .89 (.12) .18 (.23) ⬍ .51 (.15)


Gt .87 (.12) ⬎ .50 (.24) .69 (.17) ⬎ .14 (.18)
Vr .46 (.34) ⬎ .11 (.16) .06 (.20) Ø .10 (.10)
Gr .88 (.06) ⬎ .62 (.12) .59 (.07) ⬎ .08 (.08)
b .09 (.06) ⬎ .02 (.01) .19 (.06) ⬍ .40 (.19)
␣ .36 (.11) Ø .34 (.43) .53 (.04) Ø .40 (.10)
Note. Vt ⫽ verbatim memory for targets; Gt ⫽ gist memory for targets; Vr ⫽ verbatim memory for related distractors; Gr ⫽ gist memory for related
distractors; b ⫽ guessing that an item is either a target or a related distractor; ␣ ⫽ guessing “target.” ⬎ and ⬍ symbolize substantial evidence for an effect
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

of the similarity condition in one or other direction, ⫽ symbolizes evidence for a null effect, and Ø symbolizes the absence of evidence either way. Values
This document is copyrighted by the American Psychological Association or one of its allied publishers.

in parentheses are standard deviations. Statistical evidence for difference between the similarity conditions was summarized using symbols.

dunb ⫽ 1.285, 95% CI [0.687, 1.913]), with no evidence either way of target’s verbatim trace when a related distractor was presented
in the delayed test (BF10 ⫽ 0.41). Regarding Gr, the full model was reduced. Hence, gist memory was unopposed, which increased
was also the best model. There was substantial evidence that Gr false recognition of related distractors. This was especially true in
was larger in the high than in the low associative similarity the high associative similarity condition in which gist memory for
condition in both immediate (BF10 ⫽ 1.02e ⫹ 10) and delayed related distractors was high in the delayed test. Note that verbatim
tests (BF10 ⫽ 2.1e ⫹ 24), with a larger effect in delayed (Cohen’s memory for related distractors was lower (or equal in the delayed
dunb ⫽ 3.504, 95% CI [2.651, 4.453]) than immediate test (Co- test) in the low relative to the high associative similarity condition.
hen’s dunb ⫽ 2.664, 95% CI [1.925, 3.474]). In sum, in the However, participants also retrieved less frequently the gist trace
immediate test, Vr was high, especially in the high associative of a target given a related probe in the low relative to the high
similarity condition, and could oppose Gr. In the delayed test, Vr similarity condition, which explains why false recognition rate was
was reduced, leaving Gr unopposed, especially in the high asso- low even with few verbatim retrievals. Altogether, these results
ciative similarity condition in which Gr remained high. support the idea that gist memory underlies false memories in the
long term.
Discussion Finally, as predicted, true recognition declined over time, per-
formance being lower in the delayed than in the immediate rec-
Experiment 1 assessed true and false memories of low and high ognition test. Verbatim memory for targets was also reduced in the
associatively similar word lists in immediate and delayed recog- delayed test. It contributes to the same extent to true recognition of
nition tests when both WM maintenance mechanisms were avail- targets in both associative similarity conditions in the immediate
able. The classic false memory effect (e.g., Roediger & McDer- test but, unexpectedly, it was lower in the high relative to the low
mott, 1995) was replicated in the delayed test. Indeed, presenting associative similarity condition in the delayed test. This might be
lists of high associatively similar words increased false recognition because a high associative similarity between memory words
of related distractors in the recognition test performed after a fosters gist formation. Indeed, gist memory was larger in the high
2-min countdown sequence following the presentation of the last than in the low similarity condition in both tests. Gist memory for
list. However, our results did not replicate those of recent studies targets was also reduced in the delayed test. However, as expected,
showing the false memory effect in immediate test (e.g., Atkins & the effect was smaller than for verbatim memory.
Reuter-Lorenz, 2008). The false recognition rate was low and In sum, our first experiment demonstrated that the false mem-
similar in both associative similarity conditions in the recognition ories do not occur at short term when items are maintained in WM.
test performed immediately after the presentation of each list. How By contrast, the classic false memory effect was replicated in the
can we understand this last discrepancy in findings? As we ex- LTM test, showing that our material and procedure can allow the
pected, when participants can actively maintain memory items in emergence of false memories. These findings and those obtained
WM, no false memories occur. It is therefore likely that the on gist and verbatim memory suggest that, on the one hand, WM
occurrence of short-term false memories in previous experiments maintenance might prevent short-term false memories by strength-
was at least partly due to the obstruction of WM maintenance. ening verbatim memory. On the other hand, WM maintenance
Moreover, our findings support the FTT’s prediction that gist mechanisms might also promote long-term false memories by
retrieval underlies false memories. In the immediate test, verbatim fostering long-term retention of gist memory. We conducted a
memory for related distractors was high in the high associative second experiment in which both WM maintenance mechanisms
similarity condition. Indeed, participants retrieved a target’s ver- were blocked.
batim trace given a related probe in almost 50% of cases. This
would lead them to correctly identify a related probe as a related
Experiment 2
distractor. Verbatim retrieval opposed gist retrieval, which reduced
false recognition of related distractors in the high associative In Experiment 2, both refreshing and rehearsal were prevented
similarity condition. By contrast, in the delayed test, the retrieval by asking participants to read aloud and verify math equations,
FALSE MEMORY AT SHORT AND LONG TERM 1323

which is a high attention-demanding task inducing a concurrent True recognition False recognition
articulation. The impairment of WM maintenance mechanisms High similarity Low similarity
1
should weaken both verbatim and gist memory traces for related
distractors, but the effect should be larger on verbatim traces at 0.8
short term. Hence, we expected a high rate of false recognition of 0.6
related distractors in the high similarity condition in the immediate
0.4
test, replicating Atkins and Reuter-Lorenz (2008). However, in the
delayed test, verbatim traces had suffered from temporal decay and 0.2
gist memory would not be strong enough due to the impairment of
0
refreshing. Therefore, we expected that false recognition rate Immediate test Delayed test Immediate test Delayed test
would be small and would not differ as a function of associative
similarity of the word lists. Indeed, given no available gist nor Figure 4. True and false recognition accuracy as a function of time of test
verbatim memory, recognition performance should be mostly and associative similarity in Experiment 2. Error bars represent standard
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

based on guessing. There is no reason that participants in the deviations.


This document is copyrighted by the American Psychological Association or one of its allied publishers.

high associative similarity condition incorrectly identify more


often a related probe as a target on the basis of guessing than
those in the low associative similarity condition. We also ex- high associatively similar words increased false recognition rate in
pected to replicate the decrease of true recognition over time. the immediate test, but not in the delayed test.
Because both rehearsal and refreshing were prevented, true Verbatim and gist memory. Similar analyses as in Experi-
recognition, verbatim and gist memory for targets are likely to ment 1 were performed (Tables 2 and 4), and individual parameter
be reduced relative to Experiment 1. estimations were computed.4
Verbatim and gist memory for targets. Regarding Vt, the
model that included only the effect of time of test was the best
Method model. As expected, Vt was larger in the immediate than in the
Participants. Fifty-two new participants (14 males, Mage ⫽ delayed test (Cohen’s dunb ⫽ 1.942, 95% CI [1.499, 2.436]).
19.6 years, SDage ⫽ 1.9) participated for either course credits or a Regarding Gt, the full model was the best model. Substantial
cinema ticket. Half of them were randomly assigned to one con- preference for an effect of associative similarity was gathered in
dition of associative similarity. the immediate test (BF10 ⫽ 8e ⫹ 10, Cohen’s dunb ⫽ 2.816, 95%
Materials and procedure. The same procedure from Exper- CI [2.072, 3.631]) in which Gt was larger in the high associative
iment 1 was used. The only difference concerned the secondary similarity condition, and for a null effect in the delayed test
task, which was a math equation verification task adapted from (BF10 ⫽ 0.28; Table 4).
Atkins and Reuter-Lorenz (2008). After the presentation of the Verbatim and gist memory for related distractors. Regarding
memory list and a 500-ms delay, a two-operation math equation Vr, the full model was the best model. Although Vr was low in all
was displayed during 3,000 ms followed by a 500-ms delay (see the conditions, there was a substantial evidence that Vr was larger
Figure 1). Participants read aloud the equation and decided in the high relative to the low associative similarity condition in
whether it was correct or not by pressing keys. The first operation both immediate (BF10 ⫽ 5.5, Cohen’s dunb ⫽ 0.734, 95% CI
was always multiplication or division and the second was always [0.179, 1.304]) and delayed tests (BF10 ⫽ 17.15, Cohen’s dunb ⫽
addition or subtraction. This secondary task was more attentional 0.873, 95% CI [0.312, 1.453]). Regarding Gr, the full model was
demanding than the simple RT task used in Experiment 1 to the best model. Follow-up analyses provided substantial evidence
prevent participants from refreshing memory traces, and reading that Gr was larger in the high relative to the low associative
the equations aloud prevented rehearsal. similarity condition in the immediate test (BF10 ⫽ 1.89e ⫹ 6,
Cohen’s dunb ⫽ 1.936, 95% CI [1.293, 2.625]) and no preference
either way in the delayed test (BF10 ⫽ 0.64). In sum, in the
Results immediate test, Vr was low in both associative similarity condi-
Math task accuracy was high, with an average 0.82 (SD ⫽ 0.13), tions leaving Gr, which was high especially in the high similarity
assuring that participants did not favor the memory task at the cost condition, unopposed.
of poor performance in the secondary task.
Memory accuracy. Discussion
True recognition. The model that only included time of test
Experiment 2 assessed true and false recognition of low and
was the best model (see Table 2). As predicted, true recognition
high associatively similar word lists in immediate and delayed
rate was higher in the immediate than in the delayed test (Cohen’s
recognition tests when both WM maintenance mechanisms were
dunb ⫽ 1.413, 95% CI [1.042, 1.819]; Figure 4).
blocked. By preventing the maintenance of information in WM,
False recognition. The full model was the best model (see we were able to replicate in this experiment the results of Atkins
Table 2). Follow-up tests showed substantial preference for an and Reuter-Lorenz (2008) showing a false memory effect when a
effect of associative similarity in the immediate test (BF10 ⫽ 1168, few seconds separated study from test. Indeed, results showed that
Cohen’s dunb ⫽ 1.319, 95% CI [0.730, 1.937]) with a higher rate false recognition of related distractors was higher in the high
of false recognition in the high (vs. low) associative similarity
condition, and for a null effect in the delayed test (BF10 ⫽ 0.29;
see Figure 4). Thus, contrary to Experiment 1, presenting lists of 4 2
G(df ⫽12) ⫽ 7.21, p ⫽ 0.84.
1324 ABADIE AND CAMOS

Table 4
Parameter Estimates for Verbatim and Gist Memory and Guessing Processes as a Function of Time of Test and Associative Similarity
in Experiment 2

Immediate test Delayed test


Parameters High similarity Evidence for Low similarity High similarity Evidence for Low similarity

Vt .59 (.19) Ø .57 (.11) .29 (.07) Ø .34 (.14)


Gt .74 (.20) ⬎ .26 (.12) .27 (.28) ⫽ .27 (.16)
Vr 2e-6 (3.71e-6) ⬎ 0 (0) .19 (.12) ⬎ .10 (.06)
Gr .70 (.23) ⬎ .27 (.20) .22 (.09) Ø .15 (.23)
b .18 (.04) Ø .25 (.12) .48 (.12) Ø .53 (.17)
␣ .48 (.08) ⬎ .33 (.20) .71 (.07) ⬎ .60 (.10)
Note. Vt ⫽ verbatim memory for targets; Gt ⫽ gist memory for targets; Vr ⫽ verbatim memory for related distractors; Gr ⫽ gist memory for related
distractors; b ⫽ guessing that an item is either a target or a related distractor; ␣ ⫽ guessing “target”. ⬎ and ⬍ symbolize substantial evidence for an effect
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

of the similarity condition in one or other direction, ⫽ symbolizes evidence for a null effect, and Ø symbolizes the absence of evidence either way. Values
This document is copyrighted by the American Psychological Association or one of its allied publishers.

in parentheses are standard deviations. Statistical evidence for difference between the similarity conditions was summarized using symbols.

relative to the low associative similarity condition in the immedi- Experiment 3


ate test. Moreover, this experiment also showed that, in the im-
mediate test, participants were more likely to retrieve the gist than In Experiment 3, participants performed a low attention-
the verbatim trace of a target given a related distractor probe as demanding secondary task under concurrent articulation. They
retrieval cue. Hence, it seems that preventing WM maintenance were presented with the same math equations as those used in
increases the frequency of short-term false recognition of related Experiment 2, but were asked to only read them aloud without
distractors by impeding the retrieval of verbatim memory and verifying them. This procedure prevented rehearsal, but not the
facilitating the retrieval of gist memory. use of refreshing for maintaining memory words. We expected
In this experiment, no false memory effect occurred in the to observe more false recognition of related distractors in the
delayed test. Not only verbatim memory but also gist memory for high than the low associative similarity condition for both
related distractors was low in the delayed test when a related immediate and delayed tests. The use of refreshing should
distractor probe was presented. The decrease of gist memory stimulate the retrieval of both gist and verbatim traces when
observed here might be due, as we predicted, to the obstruction of confronted with related distractors. The fuzzy-processing pref-
refreshing, something that we tested in Experiment 4. In the erence principle of FTT states that adults rely by default on gist
present experiment, given that neither gist nor verbatim memory memory. Therefore, false memories should occur in the imme-
were available, participants were likely to guess when they were diate test. There should be false memories also in the delayed
confronted with a related probe. The increase of guessing param- test because verbatim memory declines more rapidly over time
eters a and b in the delayed test relative to the immediate test (see than gist memory.
online supplementary material) was in line with this suggestion. We also predicted that true recognition would be lower in the
Finally, as expected, true recognition was lower in delayed than delayed than in the immediate test. Verbatim and gist memory for
immediate test. Verbatim memory for targets was also reduced in targets should also decline over time. The effect should neverthe-
the delayed test. Gist memory was larger in the high relative to the less be smaller for gist memory that is reinforced by refreshing. As
low associative similarity condition in the immediate test, but it in previous experiments, we also expected that gist memory would
was reduced and no longer affected by associative similarity in the be better in the high relative to the low associative similarity
delayed test. True recognition, verbatim, and gist memory for condition.
targets were lower in this experiment in which both rehearsal and
refreshing were prevented relative to Experiment 1.
To conclude, the two first experiments provided clear evidence Method
that short-term false memories depend on WM maintenance. The
following experiments aimed at disentangling the role of each WM Participants. Fifty-two participants (16 males, Mage ⫽ 22.2
maintenance mechanism on the emergence of false memories in years, SDage ⫽ 3.1) who did not participate in the previous
immediate and delayed recognition tests. Because each mainte- experiments participated for either course credits or a cinema
nance mechanism has a specific role in promoting verbatim or gist ticket. Half of them were randomly assigned to one condition of
memory traces, their respective implication in WM maintenance associative similarity.
should lead to a specific pattern in the emergence of false mem- Materials and procedure. The same materials and general
ories. In Experiment 3, only rehearsal was prevented to assess the procedure were used as in Experiment 2. The only difference was
specific role of refreshing on false memories formation. Con- that participants had only to read aloud, but not to solve, math
versely, refreshing was prevented in Experiment 4 to examine the equations. This low attention-demanding task allowed participants
role of rehearsal. After presenting the results of these two exper- to refresh memory traces but not to rehearse them. As in Experi-
iments, we compared their results with those of Experiment 2 in ment 2, each math equation was displayed during 3,000 ms with a
the cross-experiment comparison section. 500-ms ISI.
FALSE MEMORY AT SHORT AND LONG TERM 1325

Results (BF10 ⫽ 3365) and the delayed tests (BF10 ⫽ 8.17), the effect
being larger in the immediate (Cohen’s dunb ⫽ 1.406, 95% CI
We verified during testing that all participants did read aloud [0.811, 2.033]) than in the delayed test (Cohen’s dunb ⫽ 0.802,
each math equation. 95% CI [0.244, 1.377]).
Memory accuracy. Verbatim and gist memory for related distractors. Vr was
True recognition. The best model was the model which in- very low in all conditions. The additive model was the best
cluded only the effect of time of test (see Table 2). As in the two model, but preferred to the model including only the associative
first experiments, true recognition rate was higher in the immediate similarity effect by a BF10 of 1.44. Vr was larger in the low than
than in the delayed test (Cohen’s dunb ⫽ 2.005, 95% CI [1.528, in the high associative similarity condition (Cohen’s dunb ⫽
2.475]; Figure 5). 1.502, 95% CI [0.899, 2.139]). But there was no substantial
False recognition. Although the additive model was the best evidence either way for the effect of time of test (see Table 2).
model, it was preferred to the full model only by a BF10 of 2.70, Regarding Gr, the additive model was the best model, but
providing no substantial evidence against the time of test x preferred to the full model only by a BF10 of 2.69. Substantial
associative similarity interaction (see Table 2). The rate of false
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

evidence supported the effect of the associative similarity in


recognition was higher in the immediate than delayed test
This document is copyrighted by the American Psychological Association or one of its allied publishers.

both the immediate (BF10 ⫽ 7.20e ⫹ 8) and the delayed tests


(Cohen’s dunb ⫽ 0.489, 95% CI [0.206, 0.783]), and in the high (BF10 ⫽ 178531), the effect being, however, larger in the
than in the low associative similarity condition (Cohen’s dunb ⫽ immediate (Cohen’s dunb ⫽ 2.420, 95% CI [1.725, 3.176]) than
1.315, 95% CI [0.727, 1.933]). As predicted, there was a in the delayed test (Cohen’s dunb ⫽ 1.749, 95% CI [1.124,
substantial preference for the effect of associative similarity in 2.414]). To summarize, Vr was very low in all the conditions
both the immediate (BF10 ⫽ 4.74) and the delayed tests leaving Gr unopposed.
(BF10 ⫽ 189), the effect being larger in the delayed (Cohen’s
dunb ⫽ 1.145, 95% CI [0.569, 1.747]) than in the immediate test
(Cohen’s dunb ⫽ 0.729, 95% CI [0.174, 1.299]; Figure 5), due Discussion
to a greater reduction of false recognition in the low than high
Experiment 3 assessed true and false recognition of low and
associative similarity condition between the immediate and the
high associatively related word lists in immediate and delayed
delayed test.
recognition tests when only refreshing was available for maintain-
Verbatim and gist memory. Similar analyses were per-
ing words in WM. As predicted, a high rate of false memories was
formed (Tables 2 and 5), and individual parameter estimations
observed in both immediate and delayed recognition tests. False
were computed.5
recognition of related distractors was higher in the high relative to
Verbatim and gist memory for targets. Regarding Vt, the
the low associative similarity condition and it was the case in both
model that included only the effect of time of test was the best
tests. Moreover, when high associative similar word lists were
model, but was preferred to the additive model only by a BF10
presented, participants relied more on gist than on verbatim mem-
of 1.60, which provided no substantial evidence against an
ory when they responded to a related distractor. This effect was
additive effect of time of test and associative similarity. Sub-
present in both recognition tests. These results suggest that refresh-
stantial evidence supported that Vt was larger in the immediate
ing reinforces gist memory, leading participants to falsely identify
than in the delayed test (Cohen’s dunb ⫽ 3.952, 95% CI [3.186,
related distractors as targets. Finally, true recognition was reduced
4.830]). A null effect of associative similarity on Vt was also
in the delayed relative to the immediate test. Verbatim and gist
supported by substantial evidence (see Table 2). Regarding Gt,
memory for targets were also reduced in the delayed test. The
the full model was the best model, but preferred to the additive
effect was, however, smaller for gist memory. Gist memory was
model only by a BF10 of 1.36 (which provided no substantial
larger in the high relative to low associative similarity condition in
evidence against the additive effect of time of test and associa-
both tests. The next experiment was designed to investigate spe-
tive similarity). Follow-up tests provided substantial evidence
cifically the role of rehearsal on false memories formation at short
for the effect of associative similarity in both the immediate
and long delays.

Experiment 4
True recognition False recognition
High similarity Low similarity
1 In Experiment 4, participants could use rehearsal to maintain the
words, but refreshing was prevented by asking participants to
0.8
perform a silent high attention-demanding secondary task. This
0.6 task was a spatial fit task with keyed responses, which is known for
impeding refreshing (e.g., Vergauwe et al., 2014).
0.4
Because rehearsal strengthens verbatim memory, we predicted
0.2 that participants would be more likely to retrieve verbatim than
0
gist traces of the targets when confronted with related distractors in
Immediate test Delayed test Immediate test Delayed test the immediate test. This should lead them to correctly reject these
distractors. Therefore, false recognition rate should be small and
Figure 5. True and false recognition accuracy as a function of time of test
and associative similarity in Experiment 3. Error bars represent standard
deviations. 5 2
G(df ⫽12) ⫽ 5.72, p ⫽ 0.93.
1326 ABADIE AND CAMOS

Table 5
Parameter Estimates for Verbatim and Gist Memory and Guessing Processes as a Function of Time of Test and Associative Similarity
in Experiment 3

Immediate test Delayed test


Parameters High similarity Evidence for Low similarity High similarity Evidence for Low similarity

Vt .74 (.004) ⫽ .76 (.04) .35 (.19) ⫽ .42 (.10)


Gt .59 (.04) ⬎ .25 (.33) .22 (.29) ⬎ .05 (.09)
Vr 1.18e-6 (1.60e-6) ⬍ .06 (.11) .02 (.05) ⬍ .10 (.07)
Gr .77 (.04) ⬎ .42 (.20) .38 (.14) ⬎ .08 (.19)
b .15 (.09) Ø .16 (.19) .42 (.12) Ø .46 (.18)
␣ .40 (.11) Ø .36 (.05) .69 (.13) ⬎ .54 (.09)
Note. Vt ⫽ verbatim memory for targets; Gt ⫽ gist memory for targets; Vr ⫽ verbatim memory for related distractors; Gr ⫽ gist memory for related
distractors; b ⫽ guessing that an item is either a target or a related distractor; ␣ ⫽ guessing “target”. ⬎ and ⬍ symbolize substantial evidence for an effect
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

of the similarity condition in one or other direction, ⫽ symbolizes evidence for a null effect, and Ø symbolizes the absence of evidence either way. Values
This document is copyrighted by the American Psychological Association or one of its allied publishers.

in parentheses are standard deviations. Statistical evidence for difference between the similarity conditions was summarized using symbols.

should not differ as a function of associative similarity in the way regarding the associative similarity effect. Moreover,
immediate test. We expected the same pattern of findings in the follow-up tests showed a substantial preference for a null effect of
delayed test (i.e., small false recognition rate and no difference as associative similarity in the immediate test (BF10 ⫽ 0.30) and no
a function of associative similarity). Indeed, verbatim memory substantial preference either way in the delayed test (BF10 ⫽ 2.53;
should decline over time, and because refreshing was impaired see Figure 6).
during maintenance intervals, participants would not benefit of False recognition. Although the best model was the model
strong gist traces at long term. that included only the effect of time of test (see Table 2), no
True recognition rate should decline over time, performance in substantial evidence was indicated either way for this effect (see
true recognition decreasing in delayed compared with immediate Figure 6).
test. The effect should be large because rehearsal does not create Verbatim and gist memory. The same analyses were per-
memory traces that persist over time. Verbatim memory and gist formed (Tables 2 and 6), individual parameter estimations were
memory for targets should also decline over time, and gist memory computed.6
should be larger in the high relative to the low associative simi- Verbatim and gist memory for targets. Regarding Vt, the
larity condition. additive model was the best model, but preferred to the model
including only the effect of time of test only by a BF10 of 1.16.
Method Substantial evidence was obtained that Vt was larger in the imme-
Participants. Fifty new participants (11 males, 39 females, diate than in the delayed test (Cohen’s dunb ⫽ 2.964, 95% CI
Mage ⫽ 21.2 years, SDage ⫽ 2.9) received either course credits or [2.352, .663]), but no substantial evidence either way was provided
a cinema ticket for participating. Half of them were randomly for the associative similarity effect. Regarding Gt, the full model
assigned to one condition of associative similarity. was the best model. Substantial evidence supported a larger Gt in
Materials and procedure. Material and procedure were the the high (vs. low) associative similarity condition in the immediate
same as in Experiment 1, except for the secondary task. Although test (BF10 ⫽ 1.075e ⫹ 19, Cohen’s dunb ⫽ 4.953, 95% CI [3.813,
the stimuli were the same as in Experiment 1, participants had to 6.079]) but no evidence either way was obtained in the delayed test
perform a spatial fit task, which is a choice RT task, in which they (BF10 ⫽ 1.04).
decided whether the horizontal line could fit into the gap between Verbatim and gist memory for related distractors. Regarding
the two dots by pressing keys as quickly and as accurately as Vr, the additive model was the best model, but preferred to the
possible. This induced a high attentional demand, that prevented model including only the effect of time of test only by a BF10 of
participants from refreshing memory traces, although rehearsal 1.32. There was substantial evidence that Vr was larger in the
was immune as the task involved nonverbal material and was immediate than the delayed test (Cohen’s dunb ⫽ 2.497, 95% CI
silently performed. [1.762, 3.375]), but no substantial evidence either way for the
associative similarity effect. Regarding Gr, the full model was the
Results best model. Substantial evidence was obtained that Gr was larger
in the high (vs. low) associative similarity condition in the imme-
Participants nicely complied with instructions with a high mean diate test (BF10 ⫽ 4.07e ⫹ 10, Cohen’s dunb ⫽ 2.859, 95% CI
accuracy in the spatial fit task of 0.75 (SD ⫽ 0.13). [2.095, 3.698]) and no preference either way in the delayed test
Memory accuracy. (BF10 ⫽ 0.49). In sum, Vr was high in the immediate test and could
True recognition. The best model was the full model (see oppose Gr when a related distractor was presented as retrieval cue.
Table 2), but preferred to the additive model only by a BF10 of
Both memory traces declined in the delayed test.
2.11. The rate of true recognition was higher in the immediate
relative to the delayed test (Cohen’s dunb ⫽ 2.552, 95% CI [2.017,
3.158]). However, no substantial evidence was provided either 6 2
G(df ⫽12) ⫽ 0.74, p ⫽ 1.
FALSE MEMORY AT SHORT AND LONG TERM 1327

True recognition False recognition vented (Experiment 2) with those obtained when only one of them
High similarity Low similarity was available.
1

0.8
Comparison Between Experiments 2 and 3: The Role
0.6 of Attentional Refreshing
0.4
Recall that, in these two experiments, the secondary task in-
0.2 duced articulatory suppression that impaired rehearsal. Experiment
3 was designed to additionally block refreshing. In the immediate
0
Immediate test Delayed test Immediate test Delayed test test, we expected that preventing refreshing would reduce both
verbatim and gist memory for related distractors, definitively
Figure 6. True and false recognition accuracy as a function of time of test knocking out verbatim memory already impacted by rehearsal
and associative similarity in Experiment 4. Error bars represent standard impairment. Therefore, there should be more false memories in the
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

deviations. immediate test when refreshing is prevented (Experiment 3). On


This document is copyrighted by the American Psychological Association or one of its allied publishers.

the other hand, in the delayed test, responses to related distractors


should be based on the retrieval of gist memory because verbatim
Discussion memory, if any remained, declines more rapidly over time. Pre-
As expected, when only rehearsal was available, there was no venting refreshing should reduce the opportunity to create gist
longer false recognition of related distractors in both immediate memory. Therefore, in the delayed test, there should be less false
and delayed recognition tests. False recognition rates were very memories when refreshing is prevented.
low in both tests and did not differ as a function of associative A first Bayesian repeated-measures ANOVA on false recogni-
similarity. In the immediate test, verbatim memory for related tion with time of test as a within-subject factor, and availability of
distractors was high and led participants to reject related distrac- refreshing and associative similarity as between-subjects factors
tors in most of the cases. Both verbatim and gist memory were showed that the full model was the best (BF10 ⫽ 4.53e ⫹ 7). To
very low in the delayed test, therefore responses to related distrac- decompose the effects, Bayesian ANOVAs were conducted sepa-
tors were based on guessing, which was high (see online supple- rately on each test. In the immediate test, the model with the main
mentary material). These results suggest that rehearsal might help effect of associative similarity only was the best (BF10 ⫽ 14,442),
to reduce short-term false memories. with more false recognition of related distractors in the high
Finally, true recognition was reduced in the delayed relative to associative similarity condition (Cohen’s dunb ⫽ 0.894, 95% CI
the immediate test. Verbatim memory for targets was also reduced [0.489, 1.295]). In the delayed test, the full model was the best
in the delayed test. Gist memory was larger in the high than in the (BF10 ⫽ 6.34). There was more false recognition in the high
low associative similarity condition in the immediate test, and it relative to the low associative similarity condition when refreshing
was reduced and was no longer affected by associative similarity was available (Experiment 3), while this difference vanished when
in the delayed test. In the next section, we compared the experi- refreshing was prevented (Experiment 2) with support for a null
ments in order to disentangle the role of each WM maintenance effect in this latter case.
mechanism on false memory formation and on true recognition. Examining the effect of refreshing on verbatim and gist
memory for related distractors, a Bayesian repeated-measures
ANOVA on Vr with time of test as within-subject factor,
Cross-Experiment Comparisons
availability of refreshing and associative similarity as between-
To examine the role of refreshing and rehearsal on false memory subjects factors indicated that the full model was the best
and on the underlying mental representations, we compared the (BF10 ⫽ 2.06e ⫹ 20). In both immediate and delayed tests, Vr
results obtained when both maintenance mechanisms were pre- was larger in the high relative to the low associative similarity

Table 6
Parameter Estimates for Verbatim and Gist Memory and Guessing Processes as a Function of Time of Test and Associative Similarity
in Experiment 4

Immediate test Delayed test


Parameters High similarity Evidence for Low similarity High similarity Evidence for Low similarity

Vt .82 (.18) Ø .89 (.04) .25 (.28) Ø .35 (.18)


Gt .92 (.02) ⬎ .39 (.15) .02 (.02) Ø .09 (.21)
Vr .32 (.05) Ø .41 (.28) .01 (.01) Ø .05 (.07)
Gr .78 (.21) ⬎ .25 (.15) .09 (.1) Ø .05 (.12)
b .09 (.10) Ø .07(.01) .61 (.13) ⬎ .40 (.13)
␣ .30 (.15) ⬍ .40 (.03) .65 (.04) ⬎ .42 (.22)
Note. Vt ⫽ verbatim memory for targets; Gt ⫽ gist memory for targets; Vr ⫽ verbatim memory for related distractors; Gr ⫽ gist memory for related
distractors; b ⫽ guessing that an item is either a target or a related distractor; ␣ ⫽ guessing “target”. ⬎ and ⬍ symbolize substantial evidence for an effect
of the similarity condition in one or other direction, ⫽ symbolizes evidence for a null effect, and Ø symbolizes the absence of evidence either way. Values
in brackets are standard deviations. Statistical evidence for difference between the similarity conditions was summarized using symbols.
1328 ABADIE AND CAMOS

condition when refreshing was impaired (Experiment 2), similarity as between-subjects factors, indicated that the full model
whereas the reverse was true when it was available (Experiment was the best (BF10 ⫽ 416199). In the immediate test, there was a
3; BF10 ⫽ 5.69, Cohen’s dunb ⫽ 0.769, 95% CI [0.369, 1.166] substantial preference for an effect of associative similarity with
in immediate test, and BF10 ⫽ 307, Cohen’s dunb ⫽ 1.205, 95% more false recognition of related distractors in the high associative
CI [0.784, 1.621] in delayed test). The full model was also the similarity condition when rehearsal was prevented (Experiment 2)
best model on Gr (BF10 ⫽ 1.35e ⫹ 39). In the immediate test, and for a null effect of associative similarity when rehearsal was
the analysis provided no evidence either way for the effect of available (Experiment 4; BF10 ⫽ 0.36). In the delayed test, false
refreshing (BF10 ⫽ 1.34). There was however a substantial recognition rate was very low in both experiments and not affected
preference for an effect of associative similarity (BF10 ⫽ either by rehearsal nor by associative similarity, the null model
3.03e ⫹ 14), Gr being larger in the high associative similarity being the best model.
condition (Cohen’s dunb ⫽ 2.037, 95% CI [1.558, 2.509]). In A Bayesian repeated-measures ANOVA conducted on Vr
the delayed test, a substantial preference supported that Gr was showed that the best model was the model that includes main
greater in the high than in the low similarity condition when effects of the three variables and interactions between availability
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

refreshing was available (Experiment 3) and no preference of rehearsal and time of test, availability of rehearsal and associa-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

either way when refreshing was prevented (Experiment 2). tive similarity, and time of test and associative similarity (BF10 ⫽
To summarize, in the immediate test, there was as much false 6.15e ⫹ 38). However, it was preferred by only a BF10 of 2.22 to
recognition of related distractors whether refreshing was avail- the model that did not include the interaction between time of test
able or not. Contrary to our expectations, preventing refreshing and associative similarity. In the immediate test, there was sub-
did not increase false recognition. By contrast, as predicted, in stantial evidence that Vr was greater when rehearsal was available
the delayed test, the false recognition effect appeared only when (BF10 ⫽ 2.15e ⫹ 19, Cohen’s dunb ⫽ 2.505, 95% CI [1.998,
refreshing was available. The pattern of results obtained on gist 3.043]), whereas the reverse was true in the delayed test (BF10 ⫽
memory for related distractors followed the one of false recog- 9.17e ⫹ 6, Cohen’s dunb ⫽ 1.344, 95% CI [0.911, 1.772]). There
nition. On the one hand, gist memory for related distractors was was also substantial evidence that Vr was larger in the high
not affected by the availability of refreshing in the immediate associative similarity condition when rehearsal was prevented (Ex-
test. On the other hand, in the delayed test, the associative
periment 2) and no substantial preference either way when re-
similarity effect (i.e., more gist memory for related distractors
hearsal was available (BF10 ⫽ 0.76). Finally, there was evidence
in the high associative similarity condition) appeared only when
for an absence of associative similarity effect in both the imme-
refreshing was available. Verbatim memory for related distrac-
diate (BF10 ⫽ 0.313) and the delayed test (BF10 ⫽ 0.388).
tors was quasi-absent in the immediate test whether refreshing
Finally, on Gr, the best model was the model with the three main
was available or not. This might explain why false recognition
effects and interactions between availability of rehearsal and time
did not increase when refreshing was prevented. Indeed, ver-
of test and between time of test and associative similarity (BF10 ⫽
batim memory was already very low when refreshing was
1.13e ⫹ 46). Substantial preference for a null effect of availability
available. It was slightly increased in the delayed test and,
of rehearsal was gathered in the immediate test (BF10 ⫽ 0.23). In
contrary to gist memory, it was lower in the high than in the low
the delayed test, evidence supported an effect of availability of
associative similarity condition when refreshing was available
whereas the reverse was true when it was impaired. Therefore, rehearsal (BF10 ⫽ 137) with more Gr when rehearsal was impaired
in the delayed test, when refreshing was available, participants (Cohen’s dunb ⫽ 0.779, 95% CI [0.374, 1.180]). Finally, substan-
were more likely to retrieve the gist than the verbatim trace of tial evidence was gathered that there was more Gr in the high (vs.
a target when they were confronted with a related distractor, low) associative similarity condition in the immediate test (BF10 ⫽
especially when lists of associatively related words were pre- 6.23e ⫹ 17, Cohen’s dunb ⫽ 2.352, 95% CI [1.858, 2.875]), but no
sented at study. This led them to falsely identify the related evidence either way in the delayed test (BF10 ⫽ 0.76).
distractors as targets. These results suggest that refreshing Presenting lists of high associatively similar words increased
increases gist memory, which can be preserved in the long term false recognition of related distractors in the immediate test, but
and increases in turn false recognition of related distractors in only when rehearsal could not be used. Gist memory for related
the delayed test. distractors remained quite high in the immediate test whether
rehearsal was available or not. However, verbatim memory for
related distractors strongly increased in the immediate test when
Comparison Between Experiments 2 and 4: The Role rehearsal was available. Verbatim memory could then oppose gist
of Articulatory Rehearsal memory. This resulted in correct rejection of related distractors.
Recall that in Experiment 4, verbal rehearsal that was pre- These findings suggest that rehearsal reduces short-term false
vented in Experiment 2 was available while refreshing was still recognition of related distractors through the retrieval of verbatim
impaired. We expected that rehearsal would suppress false memory. In the delayed test, there was no false recognition
recognition in the immediate test. This decrease on false rec- whether rehearsal was available or not. Although both verbatim
ognition should be accompanied by an increase in the retrieval and gist memory for related distractors were increased in the
of verbatim memory for related distractors. We also examined delayed test when rehearsal was prevented, both memory traces
whether gist memory for related distractors was reduced when remained quite low whether rehearsal was available or not. Hence,
rehearsal was available. in both experiments, responses to related distractors in the delayed
A Bayesian repeated-measures ANOVA with time of test as test resulted probably from guessing processes (see online supple-
within-subject factor, and availability of rehearsal and associative mentary material).
FALSE MEMORY AT SHORT AND LONG TERM 1329

General Discussion retention. This is congruent with findings of several studies show-
ing that long-term retention requires the involvement of WM (e.g.,
The present series of experiments tested a new theoretical ac- Atkinson & Shiffrin, 1968; Camos & Portrat, 2015; Loaiza &
count of the emergence of false memories at short and long term McCabe, 2013). Moreover, false recognition of related information
that integrates the FTT conception of memory with the TBRS also decreased when long-term retention of memory words was
functional description of WM. More specifically, this study ex- reduced. Indeed, we found a positive correlation (r ⫽ .38; BF10 ⫽
plored the role of WM maintenance mechanisms on the creation of 263) between true and false recognition rates in the delayed test,
false memories with the respect to three goals: (a) revealing the suggesting that true memory is needed so that false memory errors
importance of WM on the incidence of both short- and long-term can arise.
false memory, (b) examining the contribution of verbatim and gist At first sight, these findings seems to be at odds with those of
representations, and (c) disentangling the role of each WM main- Flegal et al. (2010) and Flegal and Reuter-Lorenz (2014) showing
tenance mechanism, attentional refreshing, and articulatory re- false recognition in delayed test while WM maintenance was
hearsal, in this phenomenon. To address these goals, we conducted blocked. However, the pattern of results obtained in their experi-
four experiments in which the availability of refreshing and re-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ments when comparing true and false recognition (using the dis-
hearsal was manipulated and reliance of recognition performance
This document is copyrighted by the American Psychological Association or one of its allied publishers.

criminability index) is similar to ours: False recognition rate de-


on gist and verbatim memory was assessed. Results revealed that creased when true recognition rate also decreased in the delayed
rehearsal prevents short-term false recognition of related distrac- test. True recognition in the delayed test was also higher in their
tors through the retrieval of verbatim memory for targets, whereas experiments than in ours. It is especially the case in the study of
refreshing increases long-term false recognition through the re- Flegal and Reuter-Lorenz (2014). The main difference between
trieval of gist memory for related information. We discuss these studies is that participants were given processing instructions at
results in regard of our three goals. encoding in their studies: They had to either focus their attention
on shallow (e.g., the morphology) or deep (e.g., the meaning)
The Importance of WM on the Incidence of False characteristics of the memory items. Deep processing of semantic
associates increases the rate of false recognition of related distrac-
Memory
tors and also increases accurate memory for the studied items in
Previous studies showed that false recognition of semantically the delayed test. Other studies using the DRM paradigm found
related words could occur a few seconds following study (Atkins similar effects in long-term tests (e.g., Rhodes & Anastasi, 2000;
& Reuter-Lorenz, 2008; Flegal et al., 2010; Flegal & Reuter- Thapar & McDermott, 2001; Toglia, Neuschatz, & Goodwin,
Lorenz, 2014). These results can be interpreted as demonstrating 1999). This suggests that processing instructions, especially deep
that WM can give rise to false memory at short term. Our critical processing instructions, might foster the retrieval of gist memory
analysis of these previous studies suggested that WM maintenance for targets in the long term. Retrieving the gist trace of a target
could be determinant, and that a detailed examination of the role of could lead participants to accept as studied both target and related
WM mechanisms on the incidence of false memories was needed. probes.
Contrary to previous studies, our results have shown that the Our findings were obtained using a recognition paradigm. One
occurrence of the false memory phenomenon in immediate test is should ask whether similar results would be obtained using recall
due to the obstruction of WM maintenance mechanisms. Indeed, as rather than recognition. Recall is considered as a better task to
expected, Experiment 1 demonstrated that there are no false mem- assess WM than recognition, which could be more indicative of
ories a few seconds after study when participants can actively LTM resources (Malmberg, 2008). However, previous experi-
maintain memory words during the retention interval. By contrast, ments showed false memories in both immediate recall and rec-
the short-term false recognition effect was replicated in Experi- ognition tests. Atkins and Reuter-Lorenz (2008) demonstrated that
ment 2 in which WM maintenance was impaired. Hence, the semantic recall errors are more likely to occur than other types of
incidence of false recognition in short-term tests does not mean errors in a short-term free-recall task when the retention interval
that semantic distortions originate necessarily from WM. Rather, was filled with a difficult distraction. In addition, studies using the
our results suggest that false memory errors found in short-term classic DRM paradigm (e.g., Roediger & McDermott, 1995) dem-
tests do not arise from WM, because WM prevents them, but from onstrated false memory effect using both long-term recognition
LTM. True recognition was also reduced when WM maintenance and recall tests. Hence, we expect that the current findings ob-
was blocked (Experiment 2) relative to when it was not (Experi- tained using a recognition paradigm would be replicated using a
ment 1, BF10 ⫽ 5.72 e ⫹ 7). This replicates several findings recall paradigm.
showing that impairing WM maintenance reduces short-term Finally, individual differences in WM capacity could also play
memory (STM) performance (e.g., Baddeley, 1986; Camos et al., an important role on the occurrence of false memory. Studies
2009). showed that low WM capacity individuals are more likely to recall
Furthermore, WM seems to contribute to the occurrence of intrusions than high WM capacity individuals (Unsworth &
semantic distortions in the long term. Indeed, semantically related Brewer, 2010b). Other studies indicated that individual differences
lists elicited high rate of false recognition in the delayed test of in source-monitoring abilities fully mediated the relationship be-
Experiment 1, but false memories disappeared completely in Ex- tween WM capacity and intrusions in recall (Unsworth & Brewer,
periment 2 in which WM maintenance was blocked. True recog- 2010a). Individuals low in WM capacity would be less able to
nition was also reduced in Experiment 2 relative to Experiment 1 generate appropriate cues to retrieve memory items from LTM.
in the delayed test (BF10 ⫽ 17). This might indicate that prevent- Thus, individual differences in WM capacity could also impact the
ing maintenance of memory words in WM impairs their long-term retrieval of verbatim representations, which is more resource con-
1330 ABADIE AND CAMOS

suming than gist retrieval (Abadie et al., 2013). Low WM capacity Our studies also revealed that, in the delayed test, when it is
individuals could be less able to retrieve verbatim memory, more difficult to access verbatim traces, WM still has considerable
thereby fostering in these individuals the occurrence of false influence on long-term false memories by affecting gist memory.
memory. Moreover, research also indicates that individual differ- When information could be maintained in WM (Experiment 1),
ences in WM capacity also affect the ability to maintain items in gist traces were strong relative to verbatim traces and produced
WM (e.g., Unsworth & Robison, 2015). The amount of attention false recognition of related probes. By contrast, gist retrieval was
that low WM capacity individuals can allocate to maintain items in strongly reduced (BF10 ⫽ 22.6) when WM mechanisms were
WM is weaker. Hence, it is likely that preventing WM mainte- blocked (Experiment 2). Hence, there were no false memories.
nance mechanisms, especially refreshing, is more detrimental for These findings are congruent with previous studies (e.g., Abadie
high than low WM capacity individuals (as it was previously et al., 2013, 2017) showing a decrease in verbatim retrieval when
observed in developmental differences, Barrouillet, Gavens, Ver- information retention period was disrupted by the completion of a
gauwe, Gaillard, & Camos, 2009). Future research on the relation secondary task. The present study has gone further showing that
between WM and false memory should consider individual differ- WM is crucial to prevent the emergence of short-term false mem-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ence in WM capacity. ories by strengthening verbatim memory. A second contribution of


This document is copyrighted by the American Psychological Association or one of its allied publishers.

To summarize, WM is known as the structure responsible for the the present study was to reveal the importance of WM on long-
maintenance of information at short term, affecting in turn long- term retention of gist memory. Hence, it is likely that WM rein-
term storage. The present study showed that, despite contradicting forces short-term verbatim traces that help maintaining long-term
previous conclusion that WM gives rise to false memories, WM gist memory. This idea was supported by a positive correlation
has nevertheless a determinant role in the incidence of false between short-term verbatim retrieval and long-term gist retrieval
memories, because preventing active maintenance in WM in- for related probes found when maintenance of information in WM
creases them at short term and decreases them at long term. was optimal (Experiment 1; r ⫽ .54; BF10 ⫽ 436).
Finally, we also examined representations underlying true rec-
ognition of targets. FTT studies (e.g., Brainerd et al., 1999; Brain-
Mental Representations Underlying False Memory erd & Reyna, 2002) demonstrated that verbatim and gist retrieval
both support true memory for experienced items (e.g., targets),
These experiments also allowed to determine the nature of the
either because the corresponding items are specifically recollected
representations underlying short- and long-term false memory.
(verbatim retrieval) or because the meaning of the items is familiar
According to the FTT, memory performance is based on the (gist retrieval). Our findings (Experiment 1 vs. Experiment 2)
retrieval of both verbatim and gist traces (Reyna & Brainerd, showed that WM increases immediate true memory by encourag-
1995). The mix of verbatim and gist traces leading to false memory ing both gist and verbatim retrieval (BF10 ⫽ 1.57e ⫹ 16 and 16.6,
formation depends on various factors such as retrieval cues, the respectively). In the delayed test, only gist retrieval benefitted
relative accessibility of verbatim and gist traces and forgetting from the use of WM maintenance mechanisms whereas long-term
(Brainerd & Reyna, 2002). Verbatim and gist retrieval have op- verbatim retrieval, which was also very low, did not (Experiment
posite effects on false memories for related probes (e.g., items that 1 vs. Experiment 2). This is congruent with studies showing that
preserve the meaning of experience). Gist retrieval supports false verbatim traces become inaccessible more rapidly than gist (e.g.,
memories because the meaning of an item seems familiar, but Brainerd et al., 1995).
verbatim retrieval suppresses false memories by neutralizing item In sum, in both immediate and delayed tests, gist traces promote
familiarity (Reyna & Kiernan, 1994, 1995). Another familiar find- false memories for related information, whereas verbatim traces
ing is that verbatim traces become inaccessible more rapidly than minimize the incidence of false memories. These findings repli-
gist. Studies showed that false memories are more persistent than cated other ones suggesting that gist memory is the key ingredient
true memories (Brainerd, Reyna, & Brandse, 1995; Payne, Elie, in fomenting false memories (see Reyna et al., 2016). Moreover,
Blackwell, & Neuschatz, 1996; Toglia et al., 1999). The present our study is the first to show that WM modulates also false
experiments are the first to show that the relative accessibility of memory formation by either preventing them in the immediate test
verbatim and gist traces, and hence false memory formation, are through verbatim retrieval or fomenting them in the delayed test
also modulated by WM. through gist retrieval.
Our experiments demonstrated that, in the immediate test, when
verbatim traces are still accessible and strong relative to gist traces,
The Distinct Role of Attentional Refreshing and
the ability to keep information active in WM affects the incidence
of false memories through a modulation of verbatim memory.
Articulatory Rehearsal on False Memory
When WM maintenance was optimal (Experiment 1), verbatim Based on our proposed account that integrates the FTT and the
traces were strong enough to oppose gist traces and the rate of false TBRS model, Experiments 3 and 4 tested the role of each WM
memories for related probes was consequently very low. By con- maintenance mechanism, refreshing and rehearsal, on false mem-
trast, when information could not be maintained optimally in WM ories through their potential impact on verbatim and gist represen-
because WM maintenance mechanisms were blocked (Experiment tations. Experiment 3 revealed that the availability of refreshing
2), participants could not retrieve verbatim traces even in the increases false recognition in the delayed test, with more false
immediate test. Therefore, they relied on gist memory (which was recognition of related distractors in the delayed test when partic-
also reduced relative to Experiment 1, BF10 ⫽ 22,138) to make ipants could use refreshing (Experiment 3) than when they could
recognition judgment, which increased false recognition of related not (Experiment 2). Moreover, refreshing also increased gist re-
distractors. trieval in the long-term test, gist memory for related distractors
FALSE MEMORY AT SHORT AND LONG TERM 1331

being better when refreshing was available (Experiment 3 vs. other processes, like the LOP at encoding (Craik & Lockhart,
Experiment 2). Hence, as expected, refreshing helps to preserve 1972).
gist memory over the long term. Contrary to our expectations, Hence, our findings demonstrate the specific role of each WM
refreshing did not affect the incidence of false memories in the maintenance mechanisms on false memories and enlighten the
immediate test when rehearsal is not available. We expected that underlying representations these mechanisms promote. Whereas
preventing refreshing would reduce both verbatim and gist re- refreshing increases false recognition in the delayed test through
trieval for related information, but gist memory should be less the retrieval of gist memory, rehearsal reduces false recognition in
affected (e.g., Abadie et al., 2013). Hence, false recognition of the immediate test through the retrieval of verbatim memory.
related information should have increased when refreshing was Rehearsal reduces the importance of semantic processing and
prevented. However, preventing refreshing did not increase false enhances shallow processing, and by contrast, refreshing fosters
memories in the immediate test: There was false recognition semantic processing. This last result supports the TBRS model
whether refreshing was available (Experiment 3) or not (Experi- assuming that refreshing reconstructs degraded memory traces
ment 2). Consistently, gist memory for related information was not using information from LTM (Barrouillet & Camos, 2015). Ac-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

reduced by impairing refreshing and verbatim memory was very cordingly, when refreshing is impeded upon by an attention-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

low in both experiments. Thus, false recognition did not increase demanding secondary task during the retention interval, recall
when preventing refreshing probably because verbatim memory performance should be less impacted by LTM. This is exactly what
for related information was already very low when refreshing was we obtained in the delayed test: There was less false recognition of
available. related items when refreshing was prevented. However, this result
Although attentional refreshing has attracted an increasing contradicts another WM model assuming two memory systems
amount of interest in WM, this mechanism remains rather under- and two maintenance mechanisms in WM, namely the covert
specified (see Camos et al., 2018). Our study sheds light on the retrieval model (McCabe, 2008). This model suggests that refresh-
functioning of refreshing as it reveals the nature of the represen- ing serves as a covert retrieval bringing back into WM information
tations it manipulates and how it impacts LTM. Until now, the that has been displaced in LTM. Accordingly, it predicts that when
nature of the characteristics or the type of representations that refreshing is impeded upon by a high cognitive load, recall would
refreshing promotes remains an open question. Camos et al. (2011) be dependent on LTM (Rose et al., 2014, 2015). Hence, there
suggested that refreshing prompts attention toward nonphonologi- should have been more false recognition when refreshing was
cal features of the verbal memoranda, such as semantic character- prevented. However, the reverse was found in the present study,
istics. This is congruent with previous studies suggesting that therefore providing further support to the TBRS model.
semantic processing in WM may be moderated by attention-based Besides reinforcing the dissociation between the two WM main-
factors (e.g., Loaiza, McCabe, Youngblood, Rose, & Myerson, tenance mechanisms described in the TBRS model (Camos, 2015,
2011; Rose et al., 2014; Rose & Craik, 2012). Our findings bring 2017), our findings provide also strong support to Baddeley’s
further converging evidence that attentional refreshing implicates (1986) conception that articulatory rehearsal is a determinant
semantic features by either creating or enhancing gist memory mechanism in verbal WM. Recently, it has been put forward that
traces. Moreover, because gist traces are less sensitive to loss than rehearsal may have no causal role in maintaining verbal informa-
verbatim traces (e.g., Reyna & Brainerd, 1995), they can support tion (Lewandowsky & Oberauer, 2015). Rehearsal would be either
memory retrieval after long delay of retention. Impairing refresh- an epiphenomenon consecutive with strong memory traces or even
ing in WM reduces then gist traces, explaining why refreshing also harmful to memory traces by introducing some representation-
impacts LTM (Camos & Portrat, 2015; Loaiza & McCabe, 2013). based interference. Contrary to this suggestion, the current study
Contrary to refreshing, articulatory rehearsal has a strong impact shows that articulatory rehearsal, by maintaining and promoting
on short-term false memories by reducing them in the immediate verbatim traces, has a beneficial effect on verbal maintenance. It is
test. The rate of false recognition of related distractors was lower the only WM mechanism that blocks short-term false memories by
when rehearsal could be used (Experiment 4) than when it could maintaining verbatim traces. This does not reduce the importance
not (Experiment 2). As expected, rehearsal encourages verbatim of gist traces because they are sustainable over the long term, but
retrieval in the immediate test (Experiment 4 vs. Experiment 2). they lack the precision needed by some cognitive activities (e.g.,
Verbatim traces could then oppose gist traces, thereby reducing recall information in the correct order, arithmetic problem solving,
false recognition of related information. These findings are in line reasoning, comprehension, communication). Nevertheless, the
with previous studies (Camos et al., 2013; Camos et al., 2011; shield against the incidence of false memories at short term is the
Mora & Camos, 2013) showing that rehearsal strongly emphasizes maintenance of verbatim traces through rehearsal.
surface (phonological) characteristics of memory words and seems
to reduce the importance of semantic processing in WM (Rose et
Conclusion
al., 2014, 2015). For example, Rose and colleagues showed that
participants instructed to rehearse memory words during a reten- False memory can have dramatic consequences for human be-
tion interval did not exhibit any LOP effect. In our delayed test, as ings. At long term, there are well-known tragic examples of the
predicted, rehearsal did not affect false recognition. Long-term gist impact of false memory in eye-witnesses testimony or report of
and verbatim retrievals were also not affected by the availability of child abuse. At short term, although consequences of false memory
rehearsal. This is congruent with previous studies showing that are less mediatized, there may have considerable impact on the
rehearsal has only transitory and no long-term effect (Greene, quality of our decisions or the accuracy of our solutions, because
1987; Woodward, Bjork, & Jongeward, 1973), the transfer of any distortions of the mental representations stored in WM would
information from a short-term to a long-term store depending on impair ongoing mental processes, for example when reasoning or
1332 ABADIE AND CAMOS

solving problems. Even a failure in the mere short-term mainte- Brainerd, C., Reyna, V., & Kneer, R. (1995). False-recognition reversal:
nance of information can have dramatic consequences, for exam- When similarity is distinctive. Journal of Memory and Language, 34,
ple if a nurse is told to administrate one drug and mistakenly gives 157–185. http://dx.doi.org/10.1006/jmla.1995.1008
another. This comprehensive research is the first to investigate Brainerd, C. J., Reyna, V. F., & Mojardin, A. H. (1999). Conjoint recog-
nition. Psychological Review, 106, 160 –179. http://dx.doi.org/10.1037/
thoroughly the role of WM on the occurrence of short- and
0033-295X.106.1.160
long-term false memory, testing an innovative account that inte-
Brown, G. D. A., Neath, I., & Chater, N. (2007). A temporal ratio model
grates the FTT and the TBRS model. Our four experiments dem- of memory. Psychological Review, 114, 539 –576. http://dx.doi.org/10
onstrated that WM maintenance reduces false recognition for re- .1037/0033-295X.114.3.539
lated information in immediate test and increases it in delayed test. Brown, J. (1958). Some tests of the decay theory of immediate memory.
More specifically, articulatory rehearsal prevents short-term se- The Quarterly Journal of Experimental Psychology, 10, 12–21. http://
mantic distortion by enhancing the retrieval of verbatim memory dx.doi.org/10.1080/17470215808416249
for targets. By contrast, using attentional refreshing to maintain Camos, V. (2015). Storing verbal information in working memory. Current
memory words increases long-term false memories through the Directions in Psychological Science, 24, 440 – 445. http://dx.doi.org/10
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

retrieval of gist memory for related distractors. While previous .1177/0963721415606630


This document is copyrighted by the American Psychological Association or one of its allied publishers.

studies concluded that incidence of false memories at short and Camos, V. (2017). Domain-specific versus domain-general maintenance in
working memory: Reconciliation within the time-based resource sharing
long delay of retention favors the unitary memory view, the current
model. Psychology of Learning and Motivation, 67, 135–171. http://dx
results showed that distinct processes underlie false memories
.doi.org/10.1016/bs.plm.2017.03.005
when both WM and LTM can be used relative to LTM only, which Camos, V., & Barrouillet, P. (2014). Le développement de la mémoire de
supports a dual view of memory. travail: Perspectives dans le cadre du modèle de partage temporel des
ressources. Psychologie Française, 59, 21–39. http://dx.doi.org/10
.1016/j.psfr.2012.12.003
References Camos, V., Johnson, M., Loaiza, V., Portrat, S., Souza, A., & Vergauwe,
Abadie, M., Waroquier, L., & Terrier, P. (2013). Gist memory in the E. (2018). What is attentional refreshing in working memory? Annals of
unconscious-thought effect. Psychological Science, 24, 1253–1259. the New York Academy of Sciences, 1424, 19 –32. http://dx.doi.org/10
http://dx.doi.org/10.1177/0956797612470958 .1111/nyas.13616
Abadie, M., Waroquier, L., & Terrier, P. (2017). The role of gist and Camos, V., Lagner, P., & Barrouillet, P. (2009). Two maintenance mech-
verbatim memory in complex decision making: Explaining the anisms of verbal information in working memory. Journal of Memory
unconscious-thought effect. Journal of Experimental Psychology: and Language, 61, 457– 469. http://dx.doi.org/10.1016/j.jml.2009.06
Learning, Memory, and Cognition, 43, 694 –705. http://dx.doi.org/10 .002
.1037/xlm0000336 Camos, V., Mora, G., & Barrouillet, P. (2013). Phonological similarity
Atkins, A. S., & Reuter-Lorenz, P. A. (2008). False working memories? effect in complex span task. Quarterly Journal of Experimental Psy-
Semantic distortion in a mere 4 seconds. Memory & Cognition, 36, chology: Human Experimental Psychology, 66, 1927–1950. http://dx.doi
74 – 81. http://dx.doi.org/10.3758/MC.36.1.74 .org/10.1080/17470218.2013.768275
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed Camos, V., Mora, G., & Oberauer, K. (2011). Adaptive choice between
system and its control processes. In K. W. Spen (Ed.), The psychology articulatory rehearsal and attentional refreshing in verbal working mem-
of learning and motivation: Advances in research and theory (pp. ory. Memory & Cognition, 39, 231–244. http://dx.doi.org/10.3758/
89 –195). New York, NY: Academic Press. s13421-010-0011-x
Baddeley, A. D. (1986). Working memory. Oxford, UK: Clarendon Press. Camos, V., & Portrat, S. (2015). The impact of cognitive load on delayed
Barrouillet, P., Bernardin, S., Portrat, S., Vergauwe, E., & Camos, V. recall. Psychonomic Bulletin & Review, 22, 1029 –1034. http://dx.doi
(2007). Time and cognitive load in working memory. Journal of Exper- .org/10.3758/s13423-014-0772-5
imental Psychology: Learning, Memory, and Cognition, 33, 570 –585. Coane, J. H., McBride, D. M., Raulerson, B. A., III, & Jordan, J. S. (2007).
http://dx.doi.org/10.1037/0278-7393.33.3.570 False memory in a short-term memory task. Experimental Psychology,
Barrouillet, P., & Camos, V. (2015). Working memory: Loss and recon- 54, 62–70. http://dx.doi.org/10.1027/1618-3169.54.1.62
struction. Hove, UK: Psychology Press. Cowan, N. (1995). Attention and memory: An integrated framework. New
Barrouillet, P., Gavens, N., Vergauwe, E., Gaillard, V., & Camos, V. York NY: Oxford University Press.
(2009). Working memory span development: A time-based resource- Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework
sharing model account. Developmental Psychology, 45, 477– 490. http:// for memory research. Journal of Verbal Learning and Verbal Behavior, 11,
dx.doi.org/10.1037/a0014615 671– 684. http://dx.doi.org/10.1016/S0022-5371(72)80001-X
Bonin, P., Méot, A., Ferrand, L., & Bugaïska, A. (2013). Normes Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention
d’associations verbales pour 520 mots concrets et étude de leurs rela- of words in episodic memory. Journal of Experimental Psychology:
tions avec d’ autres variables psycholinguistiques. L’Année Psy- General, 104, 268 –294. http://dx.doi.org/10.1037/0096-3445.104.3.268
chologique, 113, 63–92. http://dx.doi.org/10.4074/S0003503313001048 Cumming, G. (2014). The new statistics: why and how. Psychological
Brainerd, C. J., & Reyna, V. F. (2002). Fuzzy-trace theory and false Science, 25, 7–29.
memory. Current Directions in Psychological Science, 11, 164 –169. Deese, J. (1959). On the prediction of occurrence of particular verbal
http://dx.doi.org/10.1111/1467-8721.00192 intrusions in immediate recall. Journal of Experimental Psychology, 58,
Brainerd, C. J., & Reyna, V. F. (2005). The science of false memory. New 17–22. http://dx.doi.org/10.1037/h0046671
York, NY: Oxford University Press. http://dx.doi.org/10.1093/acprof: DeStephano, D., & LeFevre, J.-A. (2004). The role of working memory in
oso/9780195154054.001.0001 mental arithmetic. The European Journal of Cognitive Psychology, 16,
Brainerd, C. J., Reyna, V. F., & Brandse, E. (1995). Are children’s false 353–386. http://dx.doi.org/10.1080/09541440244000328
memories more persistent than their true memories? Psychological Sci- Dienes, Z. (2014). Using Bayes to get the most out of non-significant
ence, 6, 359 –364. http://dx.doi.org/10.1111/j.1467-9280.1995 results. Frontiers in Psychology, 5, 781. http://dx.doi.org/10.3389/fpsyg
.tb00526.x .2014.00781
FALSE MEMORY AT SHORT AND LONG TERM 1333

Faul, F., Erdfelder, E., Buchner, A., & Lang, A.-G. (2009). Statistical Learning, Memory, and Cognition, 37, 1258 –1263. http://dx.doi.org/10
power analyses using GⴱPower 3.1: Tests for correlation and regression .1037/a0023923
analyses. Behavior Research Methods, 41, 1149 –1160. http://dx.doi.org/ Macé, A.-L., & Caza, N. (2011). The role of articulatory suppression in
10.3758/BRM.41.4.1149 immediate false recognition. Memory, 19, 891–900. http://dx.doi.org/10
Flegal, K. E., Atkins, A. S., & Reuter-Lorenz, P. A. (2010). False mem- .1080/09658211.2011.613844
ories seconds later: The rapid and compelling onset of illusory recog- Malmberg, K. J. (2008). Recognition memory: A review of the critical
nition. Journal of Experimental Psychology: Learning, Memory, and findings and an integrated theory for relating them. Cognitive Psychol-
Cognition, 36, 1331–1338. http://dx.doi.org/10.1037/a0019903 ogy, 57, 335–384. http://dx.doi.org/10.1016/j.cogpsych.2008.02.004
Flegal, K. E., & Reuter-Lorenz, P. A. (2014). Get the gist? The effects of McCabe, D. P. (2008). The role of covert retrieval in working memory
processing depth on false recognition in short-term and long-term mem- span tasks: Evidence from delayed recall tests. Journal of Memory and
ory. Memory & Cognition, 42, 701–711. http://dx.doi.org/10.3758/ Language, 58, 480 – 494. http://dx.doi.org/10.1016/j.jml.2007.04.004
s13421-013-0391-9 McDermott, K. B. (1996). The persistence of false memories in list recall.
Fürst, A. J., & Hitch, G. J. (2000). Separate roles for executive and Journal of Memory and Language, 35, 212–230. http://dx.doi.org/10
phonological components of working memory in mental arithmetic. .1006/jmla.1996.0012
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Memory & Cognition, 28, 774 –782. http://dx.doi.org/10.3758/ Mora, G., & Camos, V. (2013). Two systems of maintenance in verbal
This document is copyrighted by the American Psychological Association or one of its allied publishers.

BF03198412 working memory: Evidence from the word length effect. PLoS ONE, 8,
Gallo, D. A. (2006). Associative illusions of memory: False memory e70026. http://dx.doi.org/10.1371/journal.pone.0070026
research in DRM and related tasks. New York, NY: Psychology Press. Nairne, J. S. (1988). A framework for interpreting recency effects in
Gallo, D. A., Roberts, M. J., & Seamon, J. G. (1997). Remembering words immediate serial recall. Memory & Cognition, 16, 343–352. http://dx
not presented in lists: Can we avoid creating false memories? Psycho- .doi.org/10.3758/BF03197045
nomic Bulletin & Review, 4, 271–276. http://dx.doi.org/10.3758/ Nairne, J. S. (1990). A feature model of immediate memory. Memory &
BF03209405 Cognition, 18, 251–269. http://dx.doi.org/10.3758/BF03213879
Gallo, D. A., & Roediger, H. L., III. (2002). Variability among word lists Oberauer, K. (2002). Access to information in working memory: Exploring
in eliciting memory illusions: Evidence for associative activation and the focus of attention. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 28, 411– 421. http://dx.doi.org/10.1037/0278-
monitoring. Journal of Memory and Language, 47, 469 – 497. http://dx
7393.28.3.411
.doi.org/10.1016/S0749-596X(02)00013-X
Payne, D. G., Elie, C. J., Blackwell, J. M., & Neuschatz, J. (1996). Memory
Greene, R. L. (1987). Effects of maintenance rehearsal on human memory.
illusions: Recalling, recognizing, and recollecting events that never
Psychological Bulletin, 102, 403– 413. http://dx.doi.org/10.1037/0033-
occurred. Journal of Memory and Language, 35, 261–285. http://dx.doi
2909.102.3.403
.org/10.1006/jmla.1996.0015
Hyde, T. S., & Jenkins, J. J. (1973). Recall for words as a function of
Peterson, L. R., & Peterson, M. J. (1959). Short-term retention of individ-
semantic, graphic, and syntactic orienting tasks. Journal of Verbal
ual verbal items. Journal of Experimental Psychology, 58, 193–198.
Learning and Verbal Behavior, 12, 471– 480. http://dx.doi.org/10.1016/
http://dx.doi.org/10.1037/h0049234
S0022-5371(73)80027-1
Reyna, V. F. (2012). A new intuitionism: Meaning, memory, and devel-
JASP team. (2018). JASP (Version 0.9) [Computer software]. Retrieved
opment in fuzzy-trace theory. Judgment and Decision Making, 7, 332–
from https://jasp-stats.org/
359.
Jeffreys, H. (1961). Theory of probability (3rd ed.). United Kingdom:
Reyna, V. F., & Brainerd, C. J. (1995). Fuzzy-trace theory: An interim
Oxford University Press.
synthesis. Learning and Individual Differences, 7, 1–75. http://dx.doi
Kass, R. E., & Raftery, A. E. (1995). Bayes Factors. Journal of the
.org/10.1016/1041-6080(95)90031-4
American Statistical Association, 90, 773–795. http://dx.doi.org/10 Reyna, V. F., Corbin, J. C., Weldon, R. B., & Brainerd, C. J. (2016). How
.1080/01621459.1995.10476572 fuzzy-trace theory predicts true and false memories for words, sentences,
Klauer, K. C. (2006). Hierarchical multinomial processing tree models: A and narratives. Journal of Applied Research in Memory & Cognition, 5,
latent-class approach. Psychometrika, 71, 7–31. 1–9. http://dx.doi.org/10.1016/j.jarmac.2015.12.003
Lakens, D. (2013). Calculating and reporting effect sizes to facilitate Reyna, V. F., & Kiernan, B. (1994). The development of gist versus
cumulative science: A practical primer for t-tests and ANOVAs. Fron- verbatim memory in sentence recognition: Effects of lexical familiarity,
tiers in Psychology, 4, 1–12. semantic content, encoding instructions, and retention interval. Devel-
Lewandowsky, S., & Oberauer, K. (2015). Rehearsal in serial recall: An opmental Psychology, 30, 178 –191. http://dx.doi.org/10.1037/0012-
unworkable solution to the nonexistent problem of decay. Psychological 1649.30.2.178
Review, 122, 674 – 699. http://dx.doi.org/10.1037/a0039684 Reyna, V. F., & Kiernan, B. (1995). Meaning, memory, and the develop-
Loaiza, V. M., & Camos, V. (2018). The role of semantic representations ment of metaphor. Metaphor and Symbolic Activity, 10, 309 –331. http://
in verbal working memory. Journal of Experimental Psychology: Learn- dx.doi.org/10.1207/s15327868ms1004_5
ing, Memory, and Cognition, 44, 863– 881. http://dx.doi.org/10.1037/ Rhodes, M. G., & Anastasi, J. S. (2000). The effects of a levels-of-
xlm0000475 processing manipulation on false recall. Psychonomic Bulletin & Re-
Loaiza, V. M., & McCabe, D. P. (2012). Temporal-contextual processing view, 7, 158 –162. http://dx.doi.org/10.3758/BF03210735
in working memory: Evidence from delayed cued recall and delayed free Robinson, K. J., & Roediger, H. L., III. (1997). Associative processes in
recall tests. Memory & Cognition, 40, 191–203. http://dx.doi.org/10 false recall and false recognition. Psychological Science, 8, 231–237.
.3758/s13421-011-0148-2 http://dx.doi.org/10.1111/j.1467-9280.1997.tb00417.x
Loaiza, V. M., & McCabe, D. P. (2013). The influence of aging on Roediger, H. L., & McDermott, K. B. (1995). Creating false memories:
attentional refreshing and articulatory rehearsal during working memory Remembering words not presented in lists. Journal of Experimental
on later episodic memory performance. Aging, Neuropsychology and Psychology, 21, 803– 814.
Cognition, 20, 37– 41. Roediger, H. L., McDermott, K. B., & Robinson, K. J. (1998). The role of
Loaiza, V. M., McCabe, D. P., Youngblood, J. L., Rose, N. S., & Myerson, associative processes in producing false remembering. In M. A. Con-
J. (2011). The influence of levels of processing on recall from working way, S. Gathercole, & C. Cornoldi (Eds.), Theories of memory II (pp.
memory and delayed recall tasks. Journal of Experimental Psychology: 187–245). Hove, UK: Psychological Press.
1334 ABADIE AND CAMOS

Rose, N. S., Buchsbaum, B. R., & Craik, F. I. M. (2014). Short-term Unsworth, N., & Brewer, G. A. (2010a). Individual differences in false
retention of a single word relies on retrieval from long-term memory recall: A latent variable analysis. Journal of Memory and Language, 62,
when both rehearsal and refreshing are disrupted. Memory & Cognition, 19 –34. http://dx.doi.org/10.1016/j.jml.2009.08.002
42, 689 –700. http://dx.doi.org/10.3758/s13421-014-0398-x Unsworth, N., & Brewer, G. A. (2010b). Variation in working memory
Rose, N. S., & Craik, F. I. M. (2012). A processing approach to the capacity and intrusions: Differences in generation or editing? The Eu-
working memory/long-term memory distinction: Evidence from the ropean Journal of Cognitive Psychology, 22, 990 –1000. http://dx.doi
levels-of-processing span task. Journal of Experimental Psychology: .org/10.1080/09541440903175086
Learning, Memory, and Cognition, 38, 1019 –1029. http://dx.doi.org/10 Unsworth, N., & Engle, R. W. (2007). The nature of individual differences
.1037/a0026976 in working memory capacity: Active maintenance in primary memory
Rose, N. S., Craik, F. I., & Buchsbaum, B. R. (2015). Levels of processing and controlled search from secondary memory. Psychological Review,
in working memory: Differential involvement of frontotemporal net- 114, 104 –132. http://dx.doi.org/10.1037/0033-295X.114.1.104
works. Journal of Cognitive Neuroscience, 27, 522–532. http://dx.doi Unsworth, N., & Robison, M. K. (2015). Individual differences in the
allocation of attention to items in working memory: Evidence from
.org/10.1162/jocn_a_00738
pupillometry. Psychonomic Bulletin & Review, 22, 757–765. http://dx
Rouder, J. N., Morey, R. D., Speckman, P. L., & Province, J. M. (2012).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

.doi.org/10.3758/s13423-014-0747-6
Default Bayes factors for ANOVA designs. Journal of Mathematical
Vergauwe, E., Barrouillet, P., & Camos, V. (2009). Visual and spatial
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Psychology, 56, 356 –374. http://dx.doi.org/10.1016/j.jmp.2012.08.001


working memory are not that dissociated after all: A time-based
Shivde, G., & Anderson, M. C. (2011). On the existence of semantic
resource-sharing account. Journal of Experimental Psychology: Learn-
working memory: Evidence for direct semantic maintenance. Journal of
ing, Memory, and Cognition, 35, 1012–1028. http://dx.doi.org/10.1037/
Experimental Psychology: Learning, Memory, and Cognition, 37, 1342– a0015859
1370. http://dx.doi.org/10.1037/a0024832 Vergauwe, E., Camos, V., & Barrouillet, P. (2014). The impact of storage
Stahl, C., & Klauer, K. C. (2008). A simplified conjoint recognition on processing: How is information maintained in working memory?
paradigm for the measurement of gist and verbatim memory. Journal of Journal of Experimental Psychology: Learning, Memory, and Cogni-
Experimental Psychology: Learning, Memory, and Cognition, 34, 570 – tion, 40, 1072–1095. http://dx.doi.org/10.1037/a0035779
586. http://dx.doi.org/10.1037/0278-7393.34.3.570 Woodward, A. E., Jr., Bjork, R. A., & Jongeward, R. H., Jr. (1973). Recall
Thapar, A., & McDermott, K. B. (2001). False recall and false recognition and recognition as a function of primary rehearsal. Journal of Verbal
induced by presentation of associated words: Effects of retention interval Learning and Verbal Behavior, 12, 608 – 617. http://dx.doi.org/10.1016/
and level of processing. Memory & Cognition, 29, 424 – 432. http://dx S0022-5371(73)80040-4
.doi.org/10.3758/BF03196393
Toglia, M. P., Neuschatz, J. S., & Goodwin, K. A. (1999). Recall accuracy Received March 15, 2018
and illusory memories: When more is less. Memory, 7, 233–256. http:// Revision received August 13, 2018
dx.doi.org/10.1080/741944069 Accepted September 19, 2018 䡲

E-Mail Notification of Your Latest Issue Online!


Would you like to know when the next issue of your favorite APA journal will be available
online? This service is now available to you. Sign up at https://my.apa.org/portal/alerts/ and you will
be notified by e-mail when issues of interest to you become available!

S-ar putea să vă placă și