Review

Neurobiology of Schemas and

Schema-Mediated Memory
Asaf Gilboa1,2,3,* and Hannah Marlatte1,2
Schemas are superordinate knowledge structures that reflect abstracted com-
Trends
monalities across multiple experiences, exerting powerful influences over how
Similar but discernible neural networks
events are perceived, interpreted, and remembered. Activated schema tem- mediate schema instantiation and
plates modulate early perceptual processing, as they get populated with spe- schematic influences on mnemonic
functions, be it enhancement of or
cific informational instances (schema instantiation). Instantiated schemas, in interference with accurate memory
turn, can enhance or distort mnemonic processing from the outset (at encod- acquisition.
ing), impact offline memory transformation and accelerate neocortical
Schema networks include the ventro-
integration. Recent studies demonstrate distinctive neurobiological processes medial prefrontal cortex (vmPFC), hip-
underlying schema-related learning. Interactions between the ventromedial pocampus, angular gyrus, and
posterior cortical regions, although
prefrontal cortex (vmPFC), hippocampus, angular gyrus (AG), and unimodal
schema instantiation may engage
associative cortices support context-relevant schema instantiation and more ventral parts of the vmPFC and
schema mnemonic effects. The vmPFC and hippocampus may compete (as more widespread cortical structures.

suggested by some models) or synchronize (as suggested by others) to opti- Schema instantiation appears to
mize schema-related learning depending on the specific operationalization of involve earlier processes than
schema memory. This highlights the need for more precise definitions of schema-mediated mnemonics and
possibly sustained activity across the
memory schemas. network.

What Is a Memory Schema? Mutual inhibition may exist between

vmPFC-centered and hippocampus-
During development humans establish large-scale knowledge systems that continue to
centered mnemonic processes under
develop throughout life [1,2]. These knowledge structures are intuitively drawn on to interpret conditions of very strong schematic
ongoing reality and to determine appropriate actions, so much so that it has been argued that influences or no prior knowledge,
children see only what they already know, projecting their thought schemas into things [1]. respectively.

Schemas also greatly influence how new information is encoded, retained, and later retrieved.
Moderate degrees of prior knowledge
The idea that cognitive schemas influence how memory is constructed and reconstructed has a may lead to vmPFC–hippocampal
long history in cognitive psychology [3]. Curiously, however, only recently has the neurobiology interaction to resolve overlapping
of memory schemas been tackled systematically [4–10]. associations and their relevant
contexts.

Schemas are higher-level knowledge structures that organize lower-level representations from
long-term memory. They serve as general-form reference templates against which new
information can be compared, binding multiple features that consistently co-occur; their
elements are nonspecific, reflecting commonalities among experiences, and have considerable
overlap and interconnectedness [1,3,10]. Importantly, schemas are dynamic structures con- 1
Rotman Research Institute, Baycrest
stantly evolving with new experiences and memories [3] through processes dubbed assimi- Health Sciences, Toronto, Canada
lation (see Glossary) and accommodation [1,10]. 2
Department of Psychology, University
of Toronto, Toronto, Canada
3
Toronto Rehabilitation Institute,
The recent neuroscience literature has used the term schema rather loosely to refer to relevant University Health Network, Toronto,
mental and neurobiological prior associative networks that influence new information process- Canada
ing [8,9], a definition that matches multiple forms of prior knowledge (Box 1). The present review
therefore adopts an inclusive approach to schema definition, also because many of these forms
*Correspondence:
of prior knowledge are important components of schemas as classically defined (Figure 1). [email protected]
These subcomponents are likely to share neurobiological processes with richer, more complex (A. Gilboa).

618 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 http://dx.doi.org/10.1016/j.tics.2017.04.013
© 2017 Elsevier Ltd. All rights reserved.
 Box 1. Types of Prior Knowledge
Glossary
Multiple forms of prior knowledge bear substantial resemblance to schemas. Certain definitions equate these forms with
Accommodation: occurs in
schemas [8,9]; others consider them the building blocks of schemas and highlight important distinctions that arise from
conjunction with assimilation;
schemas' superordinate properties [10].
involves adapting existing schemas
to take into account new related
Narratives and Event Gists information and is at the heart of the
idea that schemas are essentially
Narratives are accounts of connected events that unfold over time, often according to causal principles. Like schemas active knowledge structures.
they are organized as associative networks. Unlike schemas narratives can be encoded in a single episode rather than Angular gyrus (AG): located in the
extracted over multiple experiences and have(see Figure 1C in main text) relatively fixed, rather than adaptable, structure inferior lateral parietal lobe, posterior
and unit details [75]. Event gist is a related term referring to high-level central elements that are critical to the overall to the supramarginal gyrus; a
storyline coherence, as opposed to low-level details whose omission would not alter the event essence [23,76]. Neural heteromodal part of the TPJ
networks of narrative comprehension and gist extraction are similar and involve the dorsal (but not the ventral) mPFC, surrounding the parietal extension of
the left IFG, the posterior STS/TPJ, the bilateral anterior temporal cortex, and midline posterior cortices [75,77]. vmPFC the STS corresponding to Brodmann
recruitment during gist extraction could reflect schema reinstatement when relevant preexisting schemas serve as area (BA) 39. Within semantic
scaffolds for interpretation of the gist of specific events (e.g., for highly typical events; see Figure 1B in main text). memory the AG supports the
integration and retrieval of
Concepts and Categories supramodal complex knowledge,
such as the integration of single
concepts into large wholes,
Categories have an associative network structure of related concepts and their linked defining features that determine a
consistent with schemas. It is highly
concept’s membership of a category [26,27,78,79]. Defining features, rather than the associative relationships between
developed in humans and almost
the nodes of the network, are what determines memberships (e.g., that mammals have fur) and therefore categories
missing in other primates. The AG is
tend to be less flexible and adaptable than schemas [10], although they can hold exceptions as anomalies (e.g., the
connected almost entirely with other
naked mole-rat). Conceptual knowledge is a critical part of most schemas and they are likely to share posterior
associative areas.
neocortical structures, but the ATL and TPJ are considered conceptual knowledge hubs [26,27] and critically support
Assimilation: the active process of
the integration of new information into conceptual knowledge networks [67,68,71,80,81].
adapting new external information to
fit with internal cognitive structures or
Statistical Regularities schemas. Individuals may apply any
mental structure available to them to
This often-implicit knowledge involves the detection and extraction of recurrent patterns of stimuli and their interrelation- assimilate a new event.
ships [82–86]. Like schemas they possess an associative network structure, are based on multiple episodes, and are Concept: a mental representation
sensitive to chronological organization. Most statistical-learning studies focus on repetition within a limited perceptual denoting a class of items or entities
dimension and require a simple motor response, limiting the capacity of the information to be updated in response to that belong to a particular category;
environmental changes. However, enriching the learning context by varying stimulus characteristics and their asso- often conceived as a hierarchy with
ciated motor outputs results in dynamic knowledge structures that allow the correct interplay between specificity and superordinate, basic-level, and
flexibility [87]. Learning statistical regularities involves local computations in modality-specific cortices coupled with subordinate levels of specificity.
multidomain contributions from the MTL/hippocampus, basal ganglia, and thalamus [86]. Gist: refers to high-order elements of
a narrative or a memory that if
omitted or changed would
substantially alter the meaning of the
storyline. Note that while schemas
are typically detail-rich and highly
superordinate cognitive structures; whether the difference between these is qualitative or elaborate cognitive structures, a gist
merely quantitative is currently unknown. is a representation of a single
episode that is stripped of much of
its detailed information.
How Prior Knowledge Is Organized into Schemas in the Brain Schema instantiation: the process
Most neuroscientific studies of memory schemas focus on the way their presence influences of matching input from ongoing
the acquisition of new information. Much less direct attention has been devoted to schema reality into activated schema
‘variables’, reflecting the manner in
representations themselves, how they are acquired, stored, and retrieved or instantiated to
which activated schemas may bias
support the multitude cognitive processes for which they are required, and how they are information processing from the input
modified as a result of new learning. stream.
Schema-linked interactions
between medial prefrontal and
Acquisition medial temporal regions (SLIMM):
Controlled experimental investigation of acquisition or accommodation [1] of large-scale a competition-based model
knowledge systems is challenging because constructing these schemas can take years [2]. accounting for the roles of the
Some studies attempted schema acquisition designs to examine the influence of pretrained vmPFC and MTL in schema-
mediated mnemonic functions.
knowledge on the learning of new associations in controlled settings. Several studies exten- Schema reinstatement: the
sively trained participants on experimental stimuli (e.g. [11–13]) or followed the acquisition of process of activating and sustaining
large-scale knowledge in naturalistic academic environments [14,15]; these studies, however, a prototypical, abstracted ‘template’

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 619

 (A)

HPC

vmPFC

Context A

TPJ

MTG/STS

ATL

(B) Instanated schema

(iii) Gist (i) Schema
extracon reinststement

(iv) Schema
accomodaon

(ii) Schema serves as

template for event
interpretaon and acon
Context A

(C) Gist extracon/semanzaon:

Schema-mediated loss of detail and gist consolidaon
Single event gist memory
Single episode memory
Detailed perceptual
informaon Accomodated schema
Schemazaon and assimilaon:
Binding pernent details across similar events

Figure 1. Hypothetical Functional Neuroanatomical Model of Memory Schemas. (A) Memory schemas are coactivated long-term representations in the posterior
neocortex [e.g., retrosplenial cortex (RSPL), middle temporal gyrus (MTG)/superior temporal sulcus (STS), anterior temporal lobe (ATL), temporoparietal junction (TPJ)]

(Figure legend continued on the bottom of the next page.)

620 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8

 Box 2. Animal Studies of the constituent properties and the
Human studies on the neurobiology of schemas were inspired by animal work that either directly or indirectly addressed interrelationships that define a typical
similar questions [4–6,17,54–57]. In a series of landmark studies, rats were exhaustively trained on a set of flavor–place instance of the knowledge structure
PAs within an ‘event arena’ that was accessed from various directions [4–6,17]. The schema comprised the shared it represents.
spatial set of trained PAs. Novel schema-congruent PAs could be acquired rapidly and, surprisingly, became Statistical learning: the ability of an
hippocampal-independent within 48 h [4], much faster than typical learning, suggesting that rapid learning-induced individual to detect environmental
neocortical plasticity is possible when new information is embedded within an existing schema [4,6,17,88]. The nature regularities and extract and retain
of schematic representations within the hippocampus was highlighted by a second approach probing how related these recurring environmental
memories are represented electrophysiologically [41]. Rats learned to associate locations with rewarded and unre- patterns.
warded objects, but reward contingencies for the same objects were context dependent and no single location within a Ventromedial prefrontal cortex
context predicted reward. Using representational similarity analysis of neural activity, a hierarchical structure emerged, (vmPFC): an ill-defined swath of
with context demonstrating anticorrelated activity patterns (suggesting pattern separation) while position, valance, and cortex along the ventral medial part
item identity showed increasing similarity (overlapping) of patterns [41]. Interestingly, when new items were learned of the prefrontal cortex; typically
within previously learned contexts, the hierarchical schema structure was immediately evident, demonstrating how prior refers to the subcallosal gyrus
knowledge rapidly influences hippocampal neural code during new learning. According to this view, neural networks (BA25), the posterior medial
within the hippocampus are critical for initial schema encoding and for integrating related features of distinct memories orbitofrontal cortex (BA14, medial
into an organized context-sensitive representation. This integration involves persistent recruitment and modification of parts of BA11), and BA12 leading
widespread synaptic circuits [89] and of firing patterns [62] with the addition of new information and depends on into medial parts of BA10. It has
hippocampal dopamine receptor activation [5] and NMDA-dependent plasticity [88] at encoding. Once consolidated, mostly been described with relation
schema storage and updating relies on neocortical areas like the anterior cingulate cortex (ACC) [6] and mPFC [17,85]. to decision-making, affective
Specifically, NMDA and AMPA receptors in the ACC [6] (but not the hippocampus [88]) are required for schema processing, and extinction learning
updating and AMPA receptors for schema expression. Updating also involves the upregulation of immediate-early but more recently also as a critical
genes in prelimbic regions of the mPFC [17], which is critical for conflict resolution when incoming information is schema node in schema memory and
incongruent [85]. The consolidation process itself relies on the hippocampus supporting the rapid generation of an memory monitoring.
immature prefrontal memory trace during encoding that undergoes functional maturation with time [90]. Finally, the
orbitofrontal cortex can also support schema-based learning by linking events based on reward value [91]; thus,
different networks are involved depending on whether schemas are ‘context based’ or ‘value based’.

typically do not track the neural signature associated with the acquisition and accommodation
of the premise information itself.

Studies that have directly examined the dynamics of schema acquisition have: (i) trained
participants on specific spatial arrays of object–location paired associates (PAs) [16] inspired
by influential rodent studies of memory schema [4–6,17] (Box 2; cf. [12]); (ii) taught participants
rule-based relationships [18] based on weather-prediction tasks; or (iii) had participants learn
overlapping paired associations that allowed inference of an unlearned pair during the test
[19–22] (we discuss these in the next section).

In one of the most intensive attempts to teach experimentally controlled new schemas,
participants were followed over 9 months as they learned ten distinctive arrays of object–
location PAs. Only some of the potential locations in each array were associated with objects,
so that empty locations could later be trained in the context of old knowledge [16]. Tracking
changes in representations of the premise PA structure took place on training days 1 and 2 and
again after 3 and 6 months of training [16]. Consistent with prior research on time-

that are temporarily bound together by the ventromedial prefrontal cortex (vmPFC) to form a superordinate knowledge template. The schematic structure activated is
context sensitive: specific associative pathways (unbroken lines) biased by the vmPFC reflect context-relevant information while other potential but contextually
irrelevant associations (broken lines) are not activated or even inhibited. The hippocampus may interact with the vmPFC during schema instantiation or its activity could
reflect memory assimilation and schema adaptation processes. (B) Schemas comprise prototypical sets of variables such as general concepts (represented by circles)
and associated action scripts (represented by squares). (i) When a schema is reinstated or activated, these variables and their interrelationships serve as a template (ii)
for the interpretation of incoming information that populates the schema nodes (instantiated schema). (iii) From the onset of encoding, schemas drive the process of gist
extraction, in which inconsequential unique details or schema-incongruent components are weakened and eventually lost while connections for schema-congruent
components are strengthened and the event gist is retained. Reiterative processes of gist extraction can ultimately lead to concept abstraction. (iv) Schemas
themselves are not static: relevant schema-incongruent components can lead to a process of schema adaptation, and with repeated experiences new components
can become assimilated into the existing schema or the existing connections can shift to accommodate the new information. (C) Memory transformation processes can
lead to multiple forms of long-term representations such as gist (detail-poor) memory for a single event or a schematic representation that results from multiple similar
experiences and that can be highly detailed but not unique.

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 621

dependent trace transformation [23–25] a shift from hippocampal to vmPFC representation
was observed over the first night of training. However, by 3 months there was no evidence of
vmPFC activity; instead, the ventrolateral PFC (vlPFC) coupled with the anterior temporal lobe
(ATL) and the AG/temporoparietal junction (TPJ) was associated with retrieval of these highly
over-learned PAs. Another study trained premise information comprising simple rules studied
over 2 days of training [18]. Similar to the short delay findings from the spatial schema [16],
retrieval of the premise rules after 24 h was associated with increased activity in the vmPFC,
posterior cingulate, and higher-level sensory cortices [18]. Based on multivoxel pattern analysis
(MVPA), the AG was hypothesized to serve as a convergence zone for low-level visual features
and high-level decision rules that comprised the schema in that paradigm.

There is clearly a need for more rigorous investigation of the dynamics of schema acquisition.
The data suggest similar patterns of brain activity as those described in the rich literature on
semantization and gist extraction [23]. However, there is some evidence that more intense
exposure results in a different neural signature that involved the vlPFC and ATL rather than the
vmPFC, bearing similarity to semantic networks [26,27]. As discussed below, however,
reinstatement of even highly learned schemas strongly engages, and even requires, the
vmPFC. Perhaps when training involves repeated encoding of specific associations it results in
denotational semantic representations. Schemas, however, involve highly variable abstract
representations of both the nodes and their interrelations, and under these conditions the
vmPFC and its interactions with posterior cortices could be necessary even when no specific
episodic information is retrieved. Training protocols that capture the abstraction of commonali-
ties across variable encoding episodes in combination with previous knowledge may result in
structured, flexible symbolic representations. That is, rather than specific PAs or formal rules,
training could involve the detection of statistical regularities but also be constrained by rational
inference of relationships based on prior abstract knowledge and assumptions [2].

Schema Reinstatement and Instantiation

Cognitive and neurocognitive models suggest that schema reinstatement and instantiation
[10,28–31]: (i) are mediated by interactions between the vmPFC and posterior neocortical
structures with possible hippocampal contributions; (ii) influence processing of incoming
information; (iii) exert their influence rapidly; and (iv) are sustained within a defined context
(Figure 1). Several studies broadly support these ideas.

Human lesion studies have been central in demonstrating the critical roles of the parietal lobes
and vmPFC in memory generally and schema memory in particular [32]. For example, lesion
studies demonstrated that the right TPJ/AG is critical for extracting language-independent
schematic iconic spatial relationships from richly textured visual representations in which they
are embedded [33]. Similarly, a patient with severe perceptual deficits (simultanagnosia)
following bilateral occipital damage (sparing the TPJ/AG) was better at matching concrete
exemplars of spatial relations only when they could be directly compared with schematic iconic
representations (abstracted analog forms) but not when compared with prepositions or with
other exemplars [34]. Together these studies point to the TPJ/AG as critical for online
processing of superordinate abstracted (visual) relationships.

Deficits in processing schema-related information online have more often been observed
following vmPFC damage [35–37]. Patients with vmPFC damage and confabulation were
impaired in making simple decisions on whether words (e.g., ‘receptionist’) were associated
with common everyday schemas (e.g., ‘visit to doctor’s office’). Thus, even when memory is not
challenged vmPFC lesions impair the process of maintaining an active schema as a template for
processing incoming information [35]. Conversely, on the Deese–Roediger–McDermott false-
memory paradigm [36,37] damage to the vmPFC protects against the normal false-memory

622 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8

 (A) Schema reinstatement and instanaon
X = −48 X = −36 X=0

(B) Schema-mediated new informaon encoding

X = −46 X = 26 X = −7

Figure 2. Activation Likelihood Estimates (ALEs) of the Convergence of Activated Foci Reported in fMRI Studies of Memory Schema (P < 0.001; >100 mm3). (A) ALE
maps for contrasts that reflect schema reinstatement and instantiation derived from 132 foci from 12 experiments (Table S1 in the supplemental information online).
Significant clusters include the ventromedial prefrontal cortex (vmPFC) (left posterior medial orbitofrontal cortex, right anterior cingulate, and bilateral subcallosal gyrus),
the bilateral anterior temporal cortex, the temporoparietal junction including the angular gyrus (R > L), the bilateral hippocampus, the fusiform/parahippocampal gyrus,
the posterior cingulate, and the left posterior cingulate/retrosplenial cortex. (B) ALE maps for contrasts that reflect schema-related encoding of new information derived
from 96 foci from 15 experiments (Table S1). Significant clusters include the bilateral anterior cingulate (L > R), the left ventrolateral prefrontal cortex, superior and inferior
parietal lobules, and angular gyrus (R > L), and the right anterior hippocampus and parahippocampal gyrus.

phenomenon of recalling non-presented target lures (e.g., ‘sleep’) after encoding thematically
related word lists (e.g., dream, slumber, tired). Presumably, patients fail to construct a meta-
mnemonic thematic representation that could bias encoding and therefore target lures are not
activated and memory is paradoxically more accurate. Similar effects were found in healthy
controls following transcranial magnetic stimulation (TMS) targeting the medial PFC (mPFC)
[38].

Several fMRI studies have demonstrated similar activity patterns when schema information is
brought online (Figure 2A). In a series of studies using simple associative learning, participants
studied triads of stimuli (e.g., ABC) where only two of the possible three associations are
explicitly learned (e.g., AB, BC), such that knowledge of the third (AC) reflects associative
inference [19–21]. Spontaneous reactivation of pre-established AB pairs, and hippocampal
interactions with content-relevant neocortical modules, predicted later BC learning as well as
inferred AC knowledge and their neural signatures [19]. In a later study, representational
similarity analysis on individual A and C items before and after triad training was used to index
the reinstatement of these mini-associative networks [20]. Because A and C are never explicitly
trained together, this provides insight on schema-based memory integration [20]. Patterns
consistent with memory integration emerged in the anterior hippocampus and posterior
vmPFC but only when new, overlapping representations (BC) were acquired in the context
of well-established prior knowledge (AB). As discussed below, this is consistent with models
that focus on the vmPFC and schema functions [8,10,39,40] and with those that highlight

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 623

hippocampal contributions to schema formation through the combination of overlapping
information across episodes [9,41–44].

Neuroimaging studies have also highlighted the role of posterior cross-modal cortices such as
the AG [12,45–48] and of modality-specific cortices [21,46–48] in schema instantiation
(Figure 2A). For example, the inferior occipitotemporal cortex (‘fusiform face area’) is implicated
in schema representation when faces are central components within newly acquired mini-
schemas [21], are associated with domain expertise [14], or contain rich pre-experimental
knowledge [47]. Connectivity analyses demonstrate coactivation of these posterior neocortical
structures with the vmPFC and hippocampus.

Electrophysiological investigations have begun to uncover the temporal dynamics of schema

instantiation, complementing the anatomical observations derived from lesion and fMRI stud-
ies. The influence of thematic contexts on processing of objects is very rapid [49–51], as
predicted by schema theory. In one study, scenes with objects being handled (e.g., basketball
game) were presented when the object was congruent (basketball) or incongruent (water-
melon) with the scene [50]. Early frontally distributed event-related potentials (ERPs) (270–
600 ms) discriminated congruent and incongruent scenes, interpreted as reflecting an overlap
of perceptual (N300) and conceptual (N400) congruency effects instigated by context violation
of activated schemas. Other studies [30,51,52] have reported schema-related effects as early
as 170 ms. Cardiologists and pulmonologists performed a diagnosis-verification task by
deciding whether an image (electrocardiogram or chest radiograph) matched a specific
diagnosis (e.g., ‘arterial fibrillation’, ‘tuberculosis’). Greater domain expertise predicted
N170 amplitudes, suggesting that having an organized and accessible knowledge structure
affects early perceptual processing of complex visual stimuli [52]. Similarly, when participants
were required to activate their self-schema to endorse only personally known faces (and not
famous or unknown faces), N170 amplitudes differentiated known from unknown faces [51].
Importantly, patients with vmPFC damage did not show the N170 familiarity modulation
although it was source localized to the inferior posterior cortex (fusiform face area) [30][30].
This is consistent with the idea of rapid, top-down schema-mediated modulation of posterior
representations. These studies demonstrate very early influences of schemas on the neural
signatures of incoming information and that the vmPFC has causal influence over these
posterior modulations.

How are such early top-down influences facilitated? According to schema theories, instantia-
tion involves activating and sustaining a general ‘template’ (Figure 1) representing contextually
relevant knowledge structures and biasing information processing from the input stream to be
consistent with activated schema variables [10,28]. This should appear as protracted, tonic
effects of interaction between the vmPFC and posterior cortical regions even before a stimulus
is processed. One study found pre-stimulus sustained changes in low-frequency power (theta)
and interregional coherence between the vmPFC and bilateral inferior and lateral temporal
cortices [30], regions relevant to schema instantiation (Figure 1). Controls, but not patients with
vmPFC lesions, demonstrated decreased interregional coherence before stimulus onset, a
pattern that can support better expression of relevant neural codes during memory encoding
and retrieval [53]. Thus, the vmPFC could bias processing of stimuli in a schema-consistent
manner by maintaining tonic low-frequency desynchronization over schema-relevant posterior
neural networks.

How Schemas Influence New Memory Formation

In addition to influencing how information is processed and interpreted online, prior knowledge
also critically impacts what aspects of an experience would be encoded, what parts would be

624 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8

retained, and what would be forgotten; it can also greatly impact the extent to which retained
representations resemble the original events.

According to the schema-linked interactions between medial prefrontal and medial

temporal regions (SLIMM) model [8], the mPFC signals the congruency between existing
schematic representations and current experience, termed resonance (cf. [40] for similar ideas
on vmPFC and memory monitoring). Greater resonance leads to a shift from memory proc-
essing driven by the medial temporal lobe (MTL) to vmPFC-based memory formation. The
vmPFC is thought to activate relevant, and suppress irrelevant, schematic information to
facilitate and constrain mnemonic processing. In situations that are congruent with prior
knowledge, the vmPFC inhibits MTL-dependent binding of arbitrary event elements.

Other models, however, emphasize the central role of the hippocampus in relating new events
to prior knowledge, in addition to its role in relational binding of arbitrary associations. These
models argue that bidirectional hippocampus–mPFC interactions are required to support
efficient memory encoding and context-dependent memory retrieval [9,41]. These latter
models build on extensive animal work, predating current interest in memory schemas, that
investigated associative inference and transitive inference in rodents [54–56] and primates [57]
and that highlighted the complementary role of the hippocampal and prefrontal cortices in
constructing flexible memory representations that allow abstraction of information. Together
with more recent animal studies (Box 2), it is suggested that the hippocampus encodes
hierarchical representations of feature overlaps of closely related events while differentiating
the contexts in which these events occur. According to this view, context-defining information
is relayed to the mPFC where distinguishable contextual ‘rules’ are encoded [9]. These rules are
engaged when the same contextual information is again relayed from the hippocampus to the
mPFC and that in turn determines the activation of appropriate hippocampal representations
and suppression of context-inappropriate memories.

As we outline below, evidence for both models exists. It should be kept in mind, however, that
events are never completely congruent of completely incongruent with existing schemas, and
existing schemas may vary in strength. The current laboratory-based experimental data may
emphasize more the joint contributions of the hippocampus and mPFC, while SLIMM might
capture the interplay between them under conditions with particularly well-established, rich
schemas.

New Learning: Encoding

Prior knowledge can enhance high-order evaluative, semantic, associative binding on the one
hand and bias lower-level perceptual processing on the other hand. Depending on the
encoding conditions, these influences can lead to reduced memory accuracy or enhance
memory formation. For example, encoding information (e.g., cactus picture) in a schema-
congruent (desert) versus -incongruent (ice rink) context leads to increased hits but also to
increased false alarms when recognition involves perceptually similar items (similar cacti) [58]
(cf. [59] for related findings). Thus, schema congruency boosts overall (conceptual) memory but
undermines perceptual encoding of fine details and reduces recollection, as indexed by ERPs
[59]. Interestingly, vmPFC lesions protect against this form of false alarms, but also eliminate the
memory benefit for schema-congruent items [58]. These patterns are consistent with the
SLIMM model as they may reflect reduced pattern separation and recollection (hippocampal
functions) when items are encoded within schema-congruent contexts.

Using a similar task during fMRI, participants encoded item–context associations by rating their
congruency (e.g., umbrella–racecourse) [60]. Congruency effects for items and their associ-
ations were reported as well as parametric mPFC increases and hippocampal decreases with

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 625

increasing encoding congruency [60]. Similarly, subsequent memory effects for verbal PAs
revealed hippocampal–mPFC connectivity for subsequent non-schema- but not schema-
related item memory [61], with similar patterns of hippocampal connectivity with the superior
temporal sulcus (STS) and inferior frontal gyrus (IFG) also reported. Such evidence is consistent
with the SLIMM model by demonstrating a tradeoff between schema-driven memory enhance-
ments mediated by mPFC–cortical interactions and detail-accuracy memory mediated by
hippocampal–cortical and possibly hippocampal–mPFC coupling.

Other studies, however, suggest that – rather than a functional tradeoff – the mPFC and
hippocampus both serve (complementary) roles during schema-mediated memory formation
[9,41,62]. For example, in the associative inference paradigm described above [19,20],
increased vmPFC–hippocampus coactivation presumably reflects replay of the premise infor-
mation during encoding of new related associations, and the extent of coupling predicts
successful expression of inferred associations [9]. Thus, rather than inhibition, during formation
of overlapping associations the vmPFC exerts control that allows conflict resolution between
competing associations in the hippocampus and the creation of an integrated schematic
representation. Several studies have observed increased coupling between the vmPFC and
hippocampus during schema-related encoding [16,47].

Not all studies report mPFC activation or interactions with other structures during schema-
related encoding of new information. In a continuous recognition paradigm, to-be-recognized
objects were presented with either a congruent or an incongruent background [63]. Neither
encoding nor repetition at these very short lags revealed mPFC activation, but instead schema-
related retrieval was associated with hippocampal and posterior temporal and visual regions
[63]. Similarly, there was no reliable mPFC activation or connectivity when medical students
encoded face-diagnosis PAs before and after they acquired medical expertise [14]. It was
suggested that only when congruence with prior knowledge is perceived subjectively is the
mPFC recruited, which is not the case for face-diagnosis PAs [14]; While task demands are
probably important, it is worth noting that faces and houses are also not inherently congruent or
incongruent, yet strong mPFC effects were reported [47].

One important question is to what extent the neural effects observed during schema-related
encoding of new information (Figure 2B) are independent of schema instantiation. Some
neuroimaging analyses suggest independent contributions of prior knowledge effects and
subsequent memory effects within the same neural network [47]. Electrophysiological studies
also suggest that schema-related subsequent memory effects have spatiotemporal signatures
that differ from those of schema instantiation (see the above section on schema reinstatement
and instantiation). ERP effects of online integration during schema-related encoding were
observed later (400 ms) [64] than those associated with schema instantiation (170–270 ms)
[30,50–52]. Subsequent correct recognition was associated with greater amplitude for
schema-congruent items over frontocentral electrodes around 400 ms, which is likely to be
related to initial semantic access, and was clearly distinguishable from the later subsequent
memory effect at 650 ms that was insensitive to congruence effects [64].

New Learning: Consolidation and Retrieval

Prior knowledge also exerts its influence on the integration of new information through delayed
offline processes of consolidation, transformation, and reorganization. One of the dramatic
findings arising from landmark animal studies [4,6] (Box 2) is that novel schema-congruent
information rapidly becomes hippocampal-independent through processes of increased neo-
cortical plasticity. Historically it was believed posterior neocortical structures are incapable of
rapid consolidation because in distributed neural networks new associative connections
override old connections leading to catastrophic interference [65]. However, modified models

626 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8

demonstrate that even distributed networks can integrate new information through rapid
changes in their representational layers (as opposed to hidden layers) with little evidence
for interference [66] if that information is congruent with prior knowledge. Empirical neurobio-
logical research supports this notion and demonstrates that neocortical catastrophic interfer-
ence can be avoided only when the new associations are related but nonoverlapping [67].
Representational layers in neural networks are likely to correspond to posterior neocortical
regions, and rapid integration appears to involve lateral and inferior temporal cortices and the
TPJ [24,68,69]. Importantly, it appears that rapid neocortical consolidation is greatly enhanced
when prior knowledge is simultaneously activated with incoming information [8,70,71], leading
to synchronized activity between relevant representational nodes [30,72]. Thus, for rapid
neocortical integration of new associations to occur, other, amodal cortical hubs such as
the vmPFC or ATL may need to potentiate neocortical synchronous activity. This prior-
knowledge-dependent rapid neocortical consolidation process [66] certainly can occur with
even minimal hippocampal involvement; whether it can be hippocampal independent is not yet
known.

Much of the evidence for rapid neocortical consolidation processes in humans comes from
schema memory and sleep studies that demonstrate unique patterns of sleep-related memory
reactivation and transformation (Box 3). Several studies have investigated post-encoding offline
processes associated with schema-related information during wakefulness. In a pioneering
study, participants viewed the first half of a movie either intact or temporally scrambled and the
other half on the next day during scanning [7]. Having intact prior narrative induced greater
intersubject synchronization in the vmPFC during encoding of the second part and weaker
mPFC–MTL functional connectivity during encoding and post-encoding rest. Greater vmPFC–
hippocampal crosstalk in this case may be necessary to compensate for the poorly organized
prior knowledge to support consolidation or schema-mediated gist extraction (Figure 1B).

Box 3. Sleep and Schemas

Reactivation during sleep plays a central role in episodic memory transformation (i.e., semantization or gist extraction)
[92,93]. It appears that slow wave sleep (SWS) is critical for both initial schema formation and the integration of distinct
memories into an existing schema [92,93]. According to one model [93], during SWS slow oscillations (0.5–4 Hz) arise in
the neocortex generating widespread synchronization. During the synchronized depolarization, faster oscillations called
spindles (11–15 Hz) originate from the thalamus and spread to the neocortex, creating optimal conditions for neural
plasticity, while high-frequency ripples (100–300 Hz) occur in the hippocampus that are associated with neuronal
reactivation (‘replay’) and facilitate the transfer of information from the hippocampus to the neocortex [92]. Also during
SWS, synapses of recently potentiated neurons are thought to be downscaled to a proportion of their strength with only
the strongly potentiated interconnections surviving [94], increasing signal-to-noise ratios in relevant brain circuits. Wake-
dependent replay may also occur, tagging relevant memories for reactivation during subsequent sleep [95], and this
preselection process is likely to be influenced by currently activated schemas. This results in the reactivation of shared
elements of several memories leading to greater strengthening of their commonalities, which will then be preserved
during downscaling, than of episodic idiosyncrasies, which will be lost.

Schema and sleep studies have provided support for aspects of this model. Associating new memories with prior
knowledge was found to be essential for successful memory reactivation and schema integration via increased spindle
activity [96]. Slow oscillations and spindles are also critical for the integration of new lexical [97] and semantic [98]
information into preexisting neocortical representations, while sleep spindle density predicts accelerated hippocampal
disengagement in schema-dependent consolidation [11].

By contrast, during associative and creative thinking preexisting schemas must be disbanded to create new schemas.
This schema disintegration is thought to preferentially occur during rapid eye movement (REM) sleep, which fosters
creativity [99]. This may explain why REM sleep has been associated with a consolidation memory benefit for the
memorization of tonal melodies and is thus consistent with a Westerners schema of music but not for atonal, schema-
inconsistent melodies [100]. Thus, while the reactivation of overlapping memories may act synergistically with neural
downscaling during SWS to initially form and update various abstract networks of neural nodes into a single refined
relational representation (i.e., schema), disintegration and flexible representation of schemas could rely on REM sleep.

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 627

Others, however, found greater vmPFC–hippocampal interactions during rest when new
information was related to prior knowledge compared with unrelated associations [21]. As
described above, these discrepancies could be related to the fact that in the first study some
prior knowledge (albeit poorly connected) was present in the ‘no schema’ condition while in the
latter unrelated associations were arbitrary. Interestingly, in other domains vmPFC signals show
nonlinear relationships with psychological and behavioral measures [102–104]. The same
might be true of schema functions where vmPFC–hippocampal connectivity is weak for
new information that is completely arbitrary or for new information that is very strongly related
to prior knowledge. The interaction between the vmPFC and hippocampus may be strongest
for mid-range relatedness and reflect the reinstatement of relevant overlapping associations to
support memory integration.

Finally, very few studies have investigated the neural basis of retrieval of information encoded in
the context of prior knowledge [12,16,63,73,74]. Some of these appear to indicate the
reestablishment of thematic context to support retrieval. For example, subsequent true-
and false-memory effects for objects that were either congruent or incongruent with previously
encoded thematic scenes (bathroom, farm) demonstrated that strong schematic contexts lead
to an increase in both correct and incorrect memories. Lateral temporal cortices and the mPFC
were associated with schema-related retrieval regardless of accuracy, whereas recollection of
true memories was associated with greater activity in the hippocampus and posterior visual
cortex [74]. Others, however, argue that congruence-related differences are less pronounced
during retrieval, when all memories may begin to depend more on schematic information,
irrespective of an event’s original congruence with prior knowledge. Using a human analog of
the rodent spatial schema task [4], it was found that retrieval of schema-defining PAs was
associated with mPFC activity while schema-related new associations were associated with
greater reaction times and greater activity in frontoparietal networks, including the AG, possibly
reflecting greater top-down attention to memory [12].

Box 4. Schema Functions and Decision-Making

Both animal and human research point to the vmPFC as a central structure mediating schema representation and its
influence on new learning. The vmPFC, however, also plays a central role in value-based decision-making [101], leading
some to suggest common mechanisms across domains [103–105]. Evidence for such a link within decision-making
research comes from the observation that personal prior experience with particular rewards and punishments is
considered key to the vmPFC’s role in decision-making [106,107].

Schematic memory is a particularly strong candidate for supporting value-based decision-making and some have even
proposed that a central role of schema extraction is to optimize reinforcement-driven behaviors in dynamic environ-
ments [44]. Value-based decision-making depends on one’s prior experiences with the available options and on the
specific context in which the decision is made. Schemas are powerful cognitive structures that organize prior knowledge
and therefore could bias the process of decision-making, either consciously or preconsciously, and promote its
efficiency. In more strategic decision-making, they can provide frames of reference through which decision-makers can
represent complex strategic problems, draw analogies across similar schemas from other domains, or directly apply
previously acquired schema to a current problem. Thus, the posterior vmPFC may serve a domain-general role in
biasing relevant long-term representations in the posterior neocortex. These serve to assess the subjective relevance
and value of new information to the current context, a function that occurs rapidly and automatically. Automatic biases
often drive behavior, as reflected by phenomena such as intention superiority effects in action-based prospective
memory that involve the implicit maintenance of bias [108,109] probably mediated by the vmPFC. Similar ideas of
vmPFC biasing representations in a context-sensitive manner have been proposed with respect to its role in the
extinction of conditioned fear responses [110]. In this view, fear extinction constitutes a new learning of a conditioned
stimulus (CS)–no unconditioned stimulus (US) association and is highly context specific. The posterior vmPFC appears
to bias the expression of one set of associations over another in a context-sensitive manner based on prior subjective
experiences.

628 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8

Concluding Remarks Outstanding Questions
Are the neurocognitive systems and
Schemas exert dramatic effects on the way events are processed and integrated into memory.
neurophysiological processes that
Some of the basic neural systems and neurobiological processes that are associated with support schemas qualitatively different
these influences have been elucidated in recent years. Particularly intriguing are discoveries on from those that support the multitude
the modulation of early perceptual processing, distortion of information processing at encod- of other, related types of prior
knowledge?
ing, acceleration of neocortical integration, and changes to offline processing during sleep and
wakeful rest. Many studies, however, use relatively low-level, informationally specific knowl-
Can statistical learning on its own
edge structures that may harness some of the subprocesses recruited by general-purpose account for the development of rich,
schemas. It remains unclear whether the activation of general-form higher-order reference structured, symbolic schematic knowl-
templates simply recruits a broader compilation of these processes or is qualitatively different. edge, as suggested by some models,
or does prior knowledge need to be
Particularly lacking are insights into how schemas themselves are constructed and accom-
assumed, constraining schema
modated, as most research focuses on assimilation of new information. acquisition?

Perhaps the greatest progress was made in identifying the neural systems that support A related question is what is the role of
schema-related functions, with particular emphasis on vmPFC–hippocampal–posterior cortical abstract hierarchical knowledge in
accelerating the information learned
interactions. Here, too, many questions are left unanswered (see Outstanding Questions). For through statistical learning?
example, the vmPFC is also identified as key to many other, seemingly unrelated, cognitive
functions (e.g., decision-making, fear extinction; Box 4). It could be that the vmPFC serves a How do the temporal dynamics of
general-purpose function of biasing information processing in a context-sensitive manner. It is a activity and cross-communication
‘schema structure’ only in the sense that context-appropriate representations and actions are between neuroanatomical regions
determine schema instantiation and
determined by the accumulation of subjective experiences. Its participation in multiple memory- schema-mediated mnemonic
related domains may arise because of its role in the detection and establishment of meta- processes?
mnemonic structures that support specific-event gist extraction, inference of non-presented
relationship, superordinate organization of concepts and action plans, and so forth. What are the roles of synchronous and
desynchronous neural oscillation in
preparing the relevant neural networks
Acknowledgments for processing of schema-relevant
The authors thank Adina Levi for her assistance in the literature review and preparation of coordinates for the ALE analysis. information?
A.G. is supported by the Natural Sciences and Engineering Research Council of Canada (grant 405649) and the Canadian
Foundation for Innovation/Ontario Innovation Trust and the Heart and Stroke Foundation Centre for Stroke Recovery. Under what schema-related condi-
tions is there an inhibitory relationship
Supplemental Information between the vmPFC and hippocam-
pus and when is there interactive
Supplemental information associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.
facilitation?
tics.2017.04.013.

Is the relationship between extent of

References prior knowledge and vmPFC–hippo-
1. Piaget, J. (1952) The Origins of Intelligence in Children (Vol. 8), 11. Hennies, N. et al. (2016) Sleep spindle density predicts the effect
International Universities Press of prior knowledge on memory consolidation. J. Neurosci. 36,
campal interaction linear or nonlinear?
2. 2. Tenenbaum, J.B. et al. (2011) How to grow a mind: statistics, 3799–3810
structure, and abstraction. Science 331, 1279–1285 12. van Buuren, M. et al. (2014) Initial investigation of the effects of Can schemas support rapid neocorti-
3. Bartlett, F.C. (1932) Remembering, Cambridge University Press an experimentally learned schema on spatial associative mem- cal consolidation independently of the
ory in humans. J. Neurosci. 34, 16662–16670
4. Tse, D. et al. (2007) Schemas and memory consolidation. Sci- hippocampus or is some initial hippo-
ence 316, 76–82 13. Brod, G. et al. (2015) Differences in the neural signature of campal support always needed?
remembering schema-congruent and schema-incongruent
5. Bethus, I. et al. (2010) Dopamine and memory: modulation of the
events. Neuroimage 117, 358–366
persistence of memory for novel hippocampal NMDA receptor-
dependent paired associates. J. Neurosci. 30, 1610–1618 14. Brod, G. et al. (2016) Knowledge acquisition during exam prep-
aration improves memory and modulates memory formation. J.
6. Wang, S.H. et al. (2012) Anterior cingulate cortex in schema
Neurosci. 36, 8103–8111
assimilation and expression. Learn. Mem. 19, 315–318
15. van Kesteren, M.T. (2014) Building on prior knowledge:
7. van Kesteren, M.T. et al. (2010) Persistent schema-dependent
schema-dependent encoding processes relate to academic
hippocampal–neocortical connectivity during memory encoding
performance. J. Cogn. Neurosci. 26, 2250–2261
and postencoding rest in humans. Proc. Natl. Acad. Sci. U. S. A.
107, 7550–7555 16. Sommer, T. (2016) The emergence of knowledge and how it
supports the memory for novel related information. Cereb. Cor-
8. van Kesteren, M.T. (2012) How schema and novelty augment
tex 27, 1906–1921
memory formation. Trends Neurosci. 35, 211–219
17. Tse, D. et al. (2011) Schema-dependent gene activation and
9. Preston, A.R. and Eichenbaum, H. (2013) Interplay of hippo-
memory encoding in neocortex. Science 333, 891–895
campus and prefrontal cortex in memory. Curr. Biol. 23, R764–
R773 18. Wagner, I.C. et al. (2015) Schematic memory components
converge within angular gyrus during retrieval. Elife 4, e09668
10. Ghosh, V.E. and Gilboa, A. (2014) What is a memory schema? A
historical perspective on current neuroscience literature. Neuro-
psychologia 53, 104–114

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 629

19. Schlichting, M.L. and Preston, A.R. (2014) Memory reactivation 44. Santoro, A. et al. (2016) Memory transformation enhances rein-
during rest supports upcoming learning of related content. Proc. forcement learning in dynamic environments. J. Neurosci. 36,
Natl. Acad. Sci. U. S. A. 111, 15845–15850 12228–12242
20. Schlichting, M.L. (2015) Learning-related representational 45. de Caso, I. et al. (2017) Knowing me, knowing you: resting-state
changes reveal dissociable integration and separation signa- functional connectivity of ventromedial prefrontal cortex disso-
tures in the hippocampus and prefrontal cortex. Nat. Commun. ciates memory related to self from a familiar other. Brain Cogn.
6, 8151 113, 65–75
21. Schlichting, M.L. and Preston, A.R. (2016) Hippocampal–medial 46. Kizilirmak, J.M. et al. (2016) Neural correlates of learning from
prefrontal circuit supports memory updating during learning and induced insight: a case for reward-based episodic encoding.
post-encoding rest. Neurobiol. Learn. Mem. 134 (Pt A), 91–106 Front. Psychol. 7, 1693
22. Backus, A.R. et al. (2016) Hippocampal–prefrontal theta oscil- 47. Liu, Z.X. et al. (2016) Effects of prior-knowledge on brain acti-
lations support memory integration. Curr. Biol. 26, 450–457 vation and connectivity during associative memory encoding.
23. Winocur, G. and Moscovitch, M. (2011) Memory transformation Cereb. Cortex 27, 1991–2009
and systems consolidation. J. Int. Neuropsychol. Soc. 17, 766– 48. Reggev, N. et al. (2016) Distinct neural suppression and encod-
780 ing effects for conceptual novelty and familiarity. J. Cogn. Neu-
24. Merhav, M. et al. (2015) Not all declarative memories are created rosci. 28, 1455–1470
equal: fast mapping as a direct route to cortical declarative 49. Gronau, N. and Shachar, M. (2015) Contextual consistency
representations. Neuroimage 117, 80–92 facilitates long-term memory of perceptual detail in barely seen
25. Takashima, A. et al. (2006) Declarative memory consolidation in images. J. Exp. Psychol. Hum. Percept. Perform. 41, 1095–
humans: a prospective functional magnetic resonance imaging 1111
study. Proc. Natl. Acad. Sci. U. S. A. 103, 756–761 50. Mudrik, L. et al. (2010) ERP evidence for context congruity
26. Binder, J.R. et al. (2009) Where is the semantic system? A effects during simultaneous object-scene processing. Neuro-
critical review and meta-analysis of 120 functional neuroimaging psychologia 48, 507–517
studies. Cereb. Cortex 19, 2767–2796 51. Gilboa, A. et al. (2009) Ventromedial prefrontal cortex lesions
27. Ralph, M.A. et al. (2017) The neural and computational bases of produce early functional alterations during remote memory
semantic cognition. Nat. Rev. Neurosci. 18, 42–55 retrieval. J. Neurosci. 29, 4871–4881

28. Thorndyke, P.W. and Yekovich, F.R. (1980) A critique of 52. Rourke, L. et al. (2016) A neural marker of medical visual exper-
schema-based theories of human story memory. Poetics 9, tise: implications for training. Adv. Health Sci. Educ. Theory
23–49 Pract. 21, 953–966

29. Fruhholz, S. et al. (2011) Face recognition under ambiguous 53. Hanslmayr, S. et al. (2016) Oscillations and episodic memory:
visual stimulation: fMRI correlates of “encoding styles”. Hum. addressing the synchronization/desynchronization conundrum.
Brain Mapp. 32, 1750–1761 Trends Neurosci. 39, 16–25

30. Gilboa, A. and Moscovitch, M. (2017) Ventromedial prefrontal 54. Devito, L.M. et al. (2010) The hippocampus contributes to
cortex generates pre-stimulus theta coherence desynchroniza- memory expression during transitive inference in mice. Hippo-
tion: a schema instantiation hypothesis. Cortex 87, 16–30 campus 20, 208–217

31. Hanson, S.J. et al. (2007) Bottom-up and top-down brain func- 55. Dusek, J.A. and Eichenbaum, H. (1997) The hippocampus and
tional connectivity underlying comprehension of everyday visual memory for orderly stimulus relations. Proc. Natl. Acad. Sci. U.
action. Brain Struct. Funct. 212, 231–244 S. A. 94, 7109–7114

32. Rosenbaum, R.S. et al. (2014) Case studies continue to illumi- 56. Bunsey, M. and Eichenbaum, H. (1996) Conservation of hippo-
nate the cognitive neuroscience of memory. Ann. N. Y. Acad. campal memory function in rats and humans. Nature 379, 255–
Sci. 1316, 105–133 257

33. Amorapanth, P. et al. (2012) Language, perception, and the 57. Buckmaster, C.A. et al. (2004) Entorhinal cortex lesions disrupt
schematic representation of spatial relations. Brain Lang. 120, the relational organization of memory in monkeys. J. Neurosci.
226–236 24, 9811–9825

34. Kranjec, A. et al. (2013) Schemas reveal spatial relations to a 58. Spalding, K.N. et al. (2015) Investigating the neural correlates of
patient with simultanagnosia. Cortex 49, 1983–1988 schemas: ventromedial prefrontal cortex is necessary for normal
schematic influence on memory. J. Neurosci. 35, 15746–15751
35. Ghosh, V.E. et al. (2014) Schema representation in patients with
ventromedial PFC lesions. J. Neurosci. 34, 12057–12070 59. Sweegers, C.C. et al. (2015) Mental schemas hamper memory
storage of goal-irrelevant information. Front. Hum. Neurosci. 9,
36. Melo, B. et al. (1999) False recall and false recognition: an
629
examination of the effects of selective and combined lesions
to the medial temporal lobe/diencephalon and frontal lobe struc- 60. van Kesteren, M.T. (2013) Differential roles for medial prefrontal
tures. Cogn. Neuropsychol. 16, 343–359 and medial temporal cortices in schema-dependent encoding:
from congruent to incongruent. Neuropsychologia 51, 2352–
37. Warren, D.E. et al. (2014) False recall is reduced by damage to
2359
the ventromedial prefrontal cortex: implications for understand-
ing the neural correlates of schematic memory. J. Neurosci. 34, 61. Bein, O. et al. (2014) Prior knowledge influences on hippocam-
7677–7682 pus and medial prefrontal cortex interactions in subsequent
memory. Neuropsychologia 64, 320–330
38. Berkers, R.M. et al. (2017) Transient medial prefrontal perturba-
tion reduces false memory formation. Cortex 88, 42–52 62. McKenzie, S. et al. (2013) Learning causes reorganization of
neuronal firing patterns to represent related experiences within a
39. Brod, G. et al. (2013) The influence of prior knowledge on
hippocampal schema. J. Neurosci. 33, 10243–10256
memory: a developmental cognitive neuroscience perspective.
Front. Behav. Neurosci. 7, 139 63. McAndrews, M.P. et al. (2016) Semantic congruence affects
hippocampal response to repetition of visual associations. Neu-
40. Gilboa, A. (2010) Strategic retrieval, confabulations, and delu-
ropsychologia 90, 235–242
sions: theory and data. Cogn. Neuropsychiatry 15, 145–180
64. Packard, P.A. et al. (2017) Semantic congruence accelerates
41. McKenzie, S. et al. (2014) Hippocampal representation of
the onset of the neural signals of successful memory encoding.
related and opposing memories develop within distinct, hierar-
J. Neurosci. 37, 291–301
chically organized neural schemas. Neuron 83, 202–215
65. McClelland, J.L. et al. (1995) Why there are complementary
42. Zeithamova, D. et al. (2012) The hippocampus and inferential
learning systems in the hippocampus and neocortex: insights
reasoning: building memories to navigate future decisions.
from the successes and failures of connectionist models of
Front. Hum. Neurosci. 6, 70
learning and memory. Psychol. Rev. 102, 419–457
43. Schlichting, M.L. and Preston, A.R. (2015) Memory integration:
neural mechanisms and implications for behavior. Curr. Opin.
Behav. Sci. 1, 1–8

630 Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8

66. McClelland, J.L. (2013) Incorporating rapid neocortical learning 89. Pavlowsky, A. et al. (2016) Persistent modifications of hippo-
of new schema-consistent information into complementary campal synaptic function during remote spatial memory. Neuro-
learning systems theory. J. Exp. Psychol. Gen. 142, 1190–1210 biol. Learn. Mem. 138, 182–197
67. Merhav, M. et al. (2014) Neocortical catastrophic interference in 90. Kitamura, T. et al. (2017) Engrams and circuits crucial for sys-
healthy and amnesic adults: a paradoxical matter of time. Hip- tems consolidation of a memory. Science 356, 73–78
pocampus 24, 1653–1662 91. Farovik, A. et al. (2015) Orbitofrontal cortex encodes memories
68. Atir-Sharon, T. et al. (2015) Decoding the formation of new within value-based schemas and represents contexts that guide
semantics: MVPA investigation of rapid neocortical plasticity memory retrieval. J. Neurosci. 35, 8333–8344
during associative encoding through fast mapping. Neural Plast. 92. Inostroza, M. and Born, J. (2013) Sleep for preserving and
2015, 804385 transforming episodic memory. Annu. Rev. Neurosci. 36, 79–
69. Takashima, A. et al. (2014) Richness of information about novel 102
words influences how episodic and semantic memory networks 93. Lewis, P.A. and Durrant, S.J. (2011) Overlapping memory replay
interact during lexicalization. Neuroimage 84, 265–278 during sleep builds cognitive schemata. Trends Cogn. Sci. 15,
70. Coutanche, M.N. and Thompson-Schill, S.L. (2014) Fast map- 343–351
ping rapidly integrates information into existing memory net- 94. Tononi, G. and Cirelli, C. (2006) Sleep function and synaptic
works. J. Exp. Psychol. Gen. 143, 2296–2303 homeostasis. Sleep Med. Rev. 10, 49–62
71. Coutanche, M.N. and Thompson-Schill, S.L. (2015) Rapid con- 95. Stickgold, R. and Walker, M.P. (2013) Sleep-dependent mem-
solidation of new knowledge in adulthood via fast mapping. ory triage: evolving generalization through selective processing.
Trends Cogn. Sci. 19, 486–488 Nat. Neurosci. 16, 139–145
72. Bakker, I. et al. (2015) Changes in theta and beta oscillations as 96. Groch, S. et al. (2017) Prior knowledge is essential for the
signatures of novel word consolidation. J. Cogn. Neurosci. 27, beneficial effect of targeted memory reactivation during sleep.
1286–1297 Sci. Rep. 7, 39763
73. van Kesteren, M.T. (2010) Retrieval of associative information 97. Tamminen, J. et al. (2010) Sleep spindle activity is associated
congruent with prior knowledge is related to increased medial with the integration of new memories and existing knowledge. J.
prefrontal activity and connectivity. J. Neurosci. 30, 15888– Neurosci. 30, 14356–14360
15894
98. Tamminen, J. et al. (2013) The role of sleep spindles and slow-
74. Webb, C.E. et al. (2016) What’s the gist? The influence of wave activity in integrating new information in semantic memory.
schemas on the neural correlates underlying true and false J. Neurosci. 33, 15376–15381
memories. Neuropsychologia 93 (Pt A), 61–75
99. Landmann, N. et al. (2015) REM sleep and memory reorganiza-
75. Mar, R.A. (2011) The neural bases of social cognition and story tion: potential relevance for psychiatry and psychotherapy. Neu-
comprehension. Annu. Rev. Psychol. 62, 103–134 robiol. Learn. Mem. 122, 28–40
76. Thorndyke, P.W. (1977) Cognitive structures in comprehension 100. Durrant, S.J. et al. (2015) Schema-conformant memories are
and memory of narrative discourse. Cogn. Psychol. 9, 77–110 preferentially consolidated during REM sleep. Neurobiol. Learn.
77. Chen, J. et al. (2017) Shared memories reveal shared structure Mem. 122, 41–50
in neural activity across individuals. Nat. Neurosci. 20, 115–125 101. Kable, J.W. and Glimcher, P.W. (2009) The neurobiology of
78. Lamberts, K. and Shanks, D. (2013) Knowledge, Concepts and decision: consensus and controversy. Neuron 63, 733–745
Categories, Psychology Press 102..Lebreton, M. et al. (2015) Automatic integration of confidence in
79. Tyler, L.K. et al. (2013) Objects and categories: feature statistics the brain valuation signal. Nat Neurosci. 18, 1159–1167
and object processing in the ventral stream. J. Cogn. Neurosci. 103. Hebscher, M. et al. (2015) Memory, decision-making, and the
25, 1723–1735 ventromedial prefrontal cortex (vmPFC): the roles of subcallosal
80. Nieuwenhuis, I.L. (2011) The neocortical network representing and posterior orbitofrontal cortices in monitoring and control
associative memory reorganizes with time in a process engaging processes. Cereb. Cortex 26, 4590–4601
the anterior temporal lobe. Cereb. Cortex 22, 2622–2633 104. Hebscher, M. and Gilboa, A. (2016) A boost of confidence: the
81. Kan, I.P. et al. (2009) Contribution of prior semantic knowledge role of the ventromedial prefrontal cortex in memory, decision-
to new episodic learning in amnesia. J. Cogn. Neurosci. 21, making, and schemas. Neuropsychologia 90, 46–58
938–944 105. Euston, D.R. et al. (2012) The role of medial prefrontal cortex in
82. Durrant, S.J. et al. (2011) Sleep-dependent consolidation of memory and decision making. Neuron 76, 1057–1070
statistical learning. Neuropsychologia 49, 1322–1331 106. Fellows, L.K. (2011) Orbitofrontal contributions to value-based
83. Tobia, M.J. et al. (2012) Neural systems mediating recognition of decision making: evidence from humans with frontal lobe dam-
changes in statistical regularities. Neuroimage 63, 1730–1742 age. Ann. N. Y. Acad. Sci. 1239, 51–58
84. Vapnik, V. (2013) The Nature of Statistical Learning Theory, 107. Rudebeck, P.H. and Murray, E.A. (2014) The orbitofrontal ora-
Springer cle: cortical mechanisms for the prediction and evaluation of
85. Richards, B.A. et al. (2014) Patterns across multiple memories specific behavioral outcomes. Neuron 84, 1143–1156
are identified over time. Nat. Neurosci. 17, 981–986 108. Haynes, J.D. et al. (2007) Reading hidden intentions in the
86. Frost, R. et al. (2015) Domain generality versus modality speci- human brain. Curr. Biol. 17, 323–328
ficity: the paradox of statistical learning. Trends Cogn. Sci. 19, 109. Uretzky, S. and Gilboa, A. (2010) Knowing your lines but missing
117–125 your cue: rostral prefrontal lesions impair prospective memory
87. Roach, N.W. et al. (2017) Generalization of prior information for cue detection, but not action–intention superiority. J. Cogn.
rapid Bayesian time estimation. Proc. Natl. Acad. Sci. U. S. A. Neurosci. 22, 2745–2757
114, 412–417 110. Maren, S. et al. (2013) The contextual brain: implications for fear
88. Dragoi, G. and Tonegawa, S. (2016) Development of schemas conditioning, extinction and psychopathology. Nat. Rev. Neuro-
revealed by prior experience and NMDA receptor knock-out. sci. 14, 417–428
Elife 2, e01326

Trends in Cognitive Sciences, August 2017, Vol. 21, No. 8 631