On Staying Grounded and Avoiding Quixotic Dead Ends
Lawrence W. Barsalou
#The Author(s) 2016. This article is published with open access at Springerlink.com
Abstract The 15 articles in this special issue on The
Representation of Concepts illustrate the rich variety of theo-
retical positions and supporting research that characterize the
area. Although much agreement exists among contributors,
much disagreement exists as well, especially about the roles
of grounding and abstraction in conceptual processing. I first
review theoretical approaches raised in these articles that I
believe are Quixotic dead ends, namely, approaches that are
principled and inspired but likely to fail. In the process, I
review various theories of amodal symbols, their distortions
of grounded theories, and fallacies in the evidence used to
support them. Incorporating further contributions across arti-
cles, I then sketch a theoretical approach that I believe is likely
to be successful, which includes grounding, abstraction, flex-
ibility, explaining classic conceptual phenomena, and making
contact with real-world situations. This account further pro-
poses that (1) a key element of grounding is neural reuse, (2)
abstraction takes the forms of multimodal compression, dis-
tilled abstraction, and distributed linguistic representation (but
not amodal symbols), and (3) flexible context-dependent rep-
resentations are a hallmark of conceptual processing.
Keywords Concepts and categories .Grounded cognition .
Although large literatures in cognitive psychology and cogni-
tive science address the cognitive mechanisms underlying
conceptual representation, this special issue of Psychonomic
Bulletin & Review focuses heavily on the underlying neural
mechanisms and whether these mechanisms are consistent
with classic amodal theories or grounded cognition. Because
these issues motivate much recent research, they receive the
most attention in the special issue. As we will also see, how-
ever, traditional issues in cognitive psychology and cognitive
science receive discussion as well and are important to inte-
grate with recent neural issues in future research.
Quixotic Dead Ends in the Study of Concepts
The study of concepts offers a significant set of challenges.
First, conceptual processing proceeds largely unconsciously,
such that it is difficult to observe concepts and their effects.
Indeed, perhaps the dominant view on the relation between
concepts and consciousness is that concepts operate complete-
ly unconsciously (for a review, see Kemmerer, 2015b).
Second, as a consequence of operating unconsciously, con-
cepts are difficult to discern in content, structure, and opera-
tion. By and large, researchers must use indirect methods to
study them. Third, concepts operate acrossan unusually broad
spectrum of intelligent activities, ranging from perception, to
language and thought, to social and affective processes. How
researchers view concepts in each of these areas can vary
widely. Approaches to concepts in language, for example,
often differ considerably from approaches in nonverbal do-
mains, such as object and scene recognition. Fourth, as a result
of the preceding challenges, concepts are associated with an
unusually wide variety of theoretical proposals and positions.
Indeed, it is difficult to think of a domain characterized by so
many different views and so much disagreement (Barsalou,
2003b,2012; Margolis & Laurence, 1999; McRae & Jones,
*Lawrence W. Barsalou
Institute of Neuroscience and Psychology, University of Glasgow, 58
Hillhead Street, Glasgow G12 8QB, UK
Psychon Bull Rev
Because of these challenges, there is the potential to head
off in the wrong direction. At least many researchers in the
area believe this, often accusing each other of being complete-
ly misguided in their approach. In this spirit, I begin with
candidates for what I will refer to as Quixotic dead ends.In
my opinion, pursuing these approaches is unlikely to yield
success in understanding concepts, specifically, and cognition,
more generally. Because these approaches often are princi-
pled, inspired, idealistic, and optimistic, while being imprac-
tical and unrealistic, I refer to them as Quixotic.
I do appreciate that what one researcher perceives as a
Quixotic dead end is another’s ultimate solution. Thus, I rec-
ognize that I may be wrong about my predictions to follow.
After presenting my predictions for approaches unlikely to
succeed in this section, I turn to approaches that I believe are
more likely to succeed in the next.
Quixotic Dead Ends in Grounded Cognition
I begin with two approaches associated with grounded cogni-
tion that are unlikely to succeed as accounts of concepts (and
of cognition more generally). I refer to these approaches and
related ones later as Bgrounded,^not Bembodied,^because
Bgrounded^better captures the central focus of the general
perspective by including other forms of grounding beside em-
bodiment, such as multimodal simulation, physical situations,
and social situations (Barsalou 2008a, Barsalou 2010;Kiefer
Simple sensory-motor accounts. One possibility is that con-
ceptual knowledge is completely constructed from sensory-
motor representations, being reducible to them. A common
theme across articles in the special issue, however, is that
simple sensory-motor accounts won’tsucceed.Notasingle
author in this issue proposes or defends this perspective (even
though it often is attributed to grounded cognition researchers,
as we will see). All authors who address the issue concur that
sensory-motor mechanisms are insufficient for explaining
concepts and conceptual processing (Binder, 2016;Dove,
2016; Jamrozik, McQuire, Cardillo, & Chatterjee, 2016;
Leshinskaya & Caramazza, 2016; Reilly, Peelle, Garcia, &
Crutch, 2016;Zwaan,2016; also see Mahon, 2015).
authors argue compellingly, extensive abstraction occurs
throughout the human conceptual system, and sensory-motor
mechanisms are unlikely to explain much of it (although
sensory-motor mechanisms in the conceptual metaphor ap-
proach attempt to explain some of it; see Dove, 2016;
Jamrozik et al., 2016). If there are actually any Don
Quixotes wandering this neighborhood, they must have their
noses close to the ground.
Anti-representation accounts. Another possibility—some-
times associated with the grounded perspective because of
its emphasis on coupling the body with the environment dur-
ing perception and action—is that cognition can be explained
without the construct of representation. From this perspective,
conceptual representations are theoretical fictions that are un-
necessary for explaining cognition. One approach from this
perspective assumes that cognition can be reduced to sensory-
motor contingencies (Engel, Maye, Kurthen, & König, 2013;
O’Regan & Noë, 2001). Another focuses on laws governing
perception, action, and the body (Chemero & Turvey, 2011;
Gibson, 1979). Although proponents sometimes seem con-
vinced that these approaches will explain cognition, not ev-
eryone else is equally convinced, believing that representation
is essential for successful accounts. For one reason, represen-
tational states appear necessary for explaining how learning
can flexibly modulate the relation between stimuli and re-
sponses (Barsalou, in press-a; Chomsky, 1959;Lachman,
Lachman, & Butterfield, 1979). For another, the ability to
represent non-present states is central to what is unique and
powerful about human cognition (Donald, 1993).
Common caricatures and distortions of grounded ap-
proaches. As described later, I believe that other approaches
to grounded cognition hold promise for providing successful
accounts of conceptual processing. My optimism about these
approaches may indeed be Quixotic, but I nevertheless con-
tinue to view them as having potential. Interestingly, perhaps,
these approaches are often caricatured and distorted by critics,
turning them into, what I would agree, are Quixotic pursuits. I
address several such caricatures and distortions presented in
the special issue next.
Machery (2016) refers to grounded cognition with the
endearing moniker of Bneo-empiricism^(also see Machery,
2007). I agree that purely empiricist accounts, incorporating
no biological constraints, are unlikely to be successful.
Indeed, some grounded cognition researchers have explicitly
expressed a commitment to nativist factors (Barsalou, 1999,
2008a; Simmons & Barsalou, 2003). Many classic nativists,
such as Kant and Reid, centrally included sensory-motor pro-
cesses, such as imagery, in their accounts (Barsalou, 1999,
2008a,2010). Nevertheless, grounded cognition researchers
continue to be called neo-empiricists, even though none of us
to my knowledge explicitly rejects important native contribu-
tions. Strong biological constraints certainly exist on the cog-
nitive system, including on conceptual processing. Consistent
with Caramazza and Shelton (1998), for example, Simmons
and Barsalou (2003) argued that feature and association areas
in the brain have been shaped by evolution to anticipate im-
portant categories, such as foods, tools, and agents.
Although Mahon (2015) is not part of the special issue on
Representation of Concepts, it is highly relevant and receives significant
Psychon Bull Rev
Leshinskaya and Caramazza (2016) state with considerable
amazement that Bextreme versions^of the Bembodiment pro-
gram,^such as Barsalou (2008a) and Pulvermüller and Fadiga
(2010), continue to Bclaim that concepts are entirely reducible to
modality-specific sensory or motor representations.^Leshinskaya
and Caramazza further state that Barsalou, Simmons, Barbey, and
Wilson (2003) endorse the reduction of concepts to sensory-
motor representations, citing the following quote as evidence,
BTheories generally assume that knowledge resides in a
modular semantic system separate from episodic mem-
ory and modality-specific systems for perception, action
and emotion. These theories further assume that concep-
tual representations are amodal—unlike representations
in modality-specific systems—and operate according to
different principles. Increasingly, researchers propose
that conceptual representations are grounded in the
First, it is difficult to see how the above quote makes any
kind of reductionist claim; it simply claims that knowledge is
grounded. Second, neither Barsalou (2008a) nor Pulvermüller
and Fadiga (2010) claimed that concepts can be reduced to
sensory-motor representations. To the contrary, Barsalou
BGrounded theories are often viewed as only using
sensory-motor representations of the external world to
represent knowledge. As a result, it is argued that
grounded theories cannot represent abstract concepts
not grounded externally. Importantly, however, embodi-
ment researchers since the classic empiricists have ar-
gued that internal states such as meta-cognition and af-
fect constitute sources of knowledge no less important
than external experience.^
Throughout later sections, Barsalou (2008a) continues to de-
scribe how both language and internal states contribute to the
representation of concepts above and beyond the sensory-motor
modalities. Analogously, Pulvermüller and Fadiga (2010) never
claimed that cognition can be reduced to sensory-motor repre-
sentations but instead focused on how action supports percep-
tion and how perception-action relations contribute to language.
Second, both Barsalou and Pulvermüller clearly claim
throughout their published articles that other mechanisms besides
sensory-motor representations are essential for explaining con-
cepts. Pulvermüller (2013), for example, argued that disembodied
mechanisms in the brain’s hub regions contribute to semantic
meaning (Pulvermüller, 2012; Pulvermüller & Garagnani,
2014). Similarly, Barsalou (1999, pp. 585, 600-603) went to
considerable lengths in proposing that internal states play central
roles in conceptual processing, especially for abstract concepts
(also see Barsalou & Wiemer-Hastings, 2005). In a different vein,
Barsalou, Santos, Simmons, and Wilson (2008)reviewedapro-
gram of research showing that distributed linguistic representa-
tions play central roles in representing and processing concepts.
Given the large numbers of publications from both groups going
beyond sensory-motor mechanisms, it is difficult to understand
how Leshinskaya and Caramazza could have missed them.
In concluding their article, Leshinskaya and Caramazza make
a rather ironic proposal about how the field should be BMoving
forward.^Specifically, they propose that future research should
focus on nonsensory attributes of knowledge, such as intention,
belief, and function. To provide background for their recent
research, Leshinskaya and Caramazza cite Wilson-Mendenhall,
Simmons, Martin, and Barsalou (2013) as having previously
illustrated this approach (i.e., by establishing non-sensory-
motor attributes associated with the concepts of convince and
arithmetic). Notably, Wilson-Mendenhall et al. (p. 921) cited
four previous articles going back to Barsalou (1999)asback-
ground for the hypothesis that Bthe lexical representation for
Bconvince^is associated with much nonlinguistic semantic
content that supports meaningful understanding of the concept,
including the intentions, beliefs, internal states, affect, and ac-
tions of self and others that unfold in a spatio-temporal context.^
Using neuroimaging, Wilson-Mendehall et al. established the
neural areas that underlie this abstract non-sensory-motor con-
tent. Given that Leshinskaya and Caramazza cited this work,
their earlier claim that Barsalou (2008a) and Barsalou et al.
(2003) reduce concepts to Bmodality-specific sensory-motor
representations^is puzzling. It’s been clear for some time that
our group views abstract representations as essential.
Goldinger et al. (2016) offer a host of common misconcep-
tions about grounded approaches, warping them into Quixotic
dead ends. I address only a few of the more salient miscon-
ceptions. First, Goldinger et al. focus on embodiment in em-
bodied cognition. Although embodiment is an important
theme of grounded cognition, it is certainly not the only theme
or arguably the most important one. Most, if not all, grounded
cognition researchers would agree that embodiment is far
from sufficient for explaining cognition and believe that it
may be irrelevant during much cognitive processing
(Barsalou, 2008a,2010). The crux of the grounded approach
is understanding how the modalities, the physical environ-
ment, the social environment, and the body contribute to cog-
nition, playing central roles in the diverse forms it takes
(Barsalou, Breazeal, & Smith, 2007).
Second, Goldinger et al. state that Bmodeling is currently
impossible in EC [embodied cognition].^Without a doubt, any
approach that can’t be modeled is a non-starter. Review articles,
such as Barsalou (2008, pp. 634-635), however, often have cited
computational models of grounded cognition, which continue to
accumulate (Adams et al. 2014; Blouw, Solodkin, Thagard, &
Eliasmith, 2015; Caligiore, Borghi, Parisi, & Baldassarre, 2010;
Eliasmith, 2013; Pulvermüller & Garagnani, 2014; Schrodt,
Layher, Neumann, & Butz, 2015; Thagard & Stewart, 2011).
Psychon Bull Rev
Given the existing literature, how could anyone responsibly
claim that computational models aren’tpossible?
Goldinger et al. further propose that grounded cognition is
not sufficiently specified to produce concrete models.
Notably, all other approaches to cognitive theory must address
this issue as well, including the classic information processing
perspective, neural nets, Bayesian modeling, and so forth. For
each of these general approaches, it is necessary to make as-
sumptions that constrain the space of models, thereby making
the construction of specific models possible. Information pro-
cessing, for example, is so general that a huge space of models
exists, many of which are implausible in humans, such as
many models in machine learning, other areas of computer
science, and mathematics. Only when assumptions relevant
to human information processing are made, does it become
possible to construct meaningful models about human cogni-
tion. Grounded cognition is no exception, with the models
developed so far reflecting various theoretical commitments
that made them possible.
Third, Goldinger et al. suggest that grounded cognition
aims to replace standard theories of cognition but is unable
to do so. In my experience, re-explaining all basic cognitive
phenomena is not something to which grounded cognition
researchers aspire at all. Again, most, if not all, grounded
cognition researchers would agree that grounded cognition,
on its own, is far from capable of explaining everything that
we know about cognition. To the contrary, grounded cognition
complements traditional approaches. Rather than aiming to
replace traditional approaches, grounded cognition develops
their relations with the modalities, the body, and the environ-
ment (Barsalou, 2008a, p. 635-636). Similarly, an integration
of grounded approaches with classic symbolic and statistical
approaches is likely and necessary (Barsalou, 1999,p.652;
Finally, Goldinger et al. present a long list of basic cogni-
tive phenomena, such as word frequency, short-term memory
scanning,and serial recall, arguing that the grounded approach
hasn’t informed our understanding of them. Goldinger et al.’s
surprisingly lengthy (and elementary) descriptions of these
phenomena give the Bcopious theoretical writing^of ground-
ed cognition a run for its money. No grounded cognition re-
searcher would argue that grounded cognition should or could
re-explain all cognitive phenomena. Again, grounded cogni-
tion aims to complement and incorporate theoretical con-
structs and empirical findings in the field, such as those on
Goldinger et al.’s arbitrary list.
In summary, I completely agree that the approaches to
grounded cognition covered in this section are dead ends,
including approaches that have no biological constraints, ap-
proaches that assume concepts are reducible to sensory-motor
systems, approaches that focus completely on the body, and
approaches that aim to replace existing cognitive theories rath-
er than complement them.
Quixotic Dead Ends Associated with Amodal
I turn next to another class of theories that I believe are equally
unlikely to succeed, namely, theories proposing that amodal
symbols constitute the central form of conceptual representa-
tion. I will assume that the amodal symbols discussed in this
section have two properties: (1) they are arbitrarily related to
their corresponding categories in the world and experience;
and (2) they can stand alone without grounding to perform
the basic computations underlying conceptual processing. In
a later section, I entertain other types of abstract representa-
tions having different properties that strike me as more
Machery’s(2016)offloading hypothesis.According to this
account, concepts consist of amodal symbols, and conceptual
processing results from operating on them. When it is
necessary to relate these results to the world, the amodal
system offloads them to sensory-motor systems, which then
perform various tasks. Perhaps this is a way to build a robot
(although not the smarter and more interesting robots that
roboticists have been building for the past 30 years).
Nevertheless, I’m betting that this is not how the human brain
sections, the offloading hypothesis strikes me as the epitome
of a Quixotic dead end.
Leshinskaya and Caramazza’s(2016)interaction
hypothesis.Similar to the offloading hypothesis,
Leshinskaya and Caramazza make a clear distinction between
concepts and sensory-motor processing, with some brain areas
supporting concepts, and others supporting sensory-motor
processing. Like Machery’s offloading hypothesis, conceptual
processing interacts with sensory-motor processing when nec-
essary, but doesn’t depend on it in any crucial ways. In other
words, the representations in these conceptual areas (presum-
ably amodal symbols) are again sufficient for stand-alone con-
ceptual processing. When processing color features in con-
cepts (e.g., yellow for banana), a brain region that represents
conceptual information about color becomes active, but no
use of visual color areas is necessary. Conversely, when color
is perceived in visual areas, it causes the retrieval of amodal
representations of color in the related conceptual area, which
can represent color without use of the visual areas. Because
this approach assumes that conceptual processing areas only
interact with sensory-motor areas when necessary, it is essen-
tially no different than Machery’s offloading hypothesis. The
clear implication of this proposal is that different brain areas
underlie concepts and sensory-motor processing, contrary to
the basic proposal of grounded cognition that systems used
in sensory-motor processing are also used to some extent
Psychon Bull Rev
during conceptual processing (as described later in the section
on neural reuse).
Interestingly, the interaction hypothesis often (but not nec-
essarily) assumes that conceptual areas lie near related
sensory-motor areas. To explain why conceptual processing
often activates regions within sensory-motor processing
streams, the interaction hypothesis proposes that conceptual
processing areas utilize amodal symbols that reside within
these streams. When processing color conceptually activates
areas in the color perception pathway (Hsu, Frankland, &
Thompson-Schill, 2012; Simmons et al., 2007), the interac-
tion hypothesis proposes that these activated areas contain
amodal symbols. In this way, it becomes possible to argue that
activations for conceptual processing in sensory-motor
streams don’t actually offer support for grounded views.
Instead, these activated areas implement conceptual process-
ing using amodal symbols that interact with nearby sensory-
motor areas when necessary.
Mahon's (2015)default explanation.Like Machery (2016)
and Leshinskaya and Caramazza (2016), Mahon (2015)as-
sumes that there is a clear distinction between conceptual
areas and sensory-motor areas, with conceptual areas utilizing
amodal symbols. Indeed, he concludes that,
BThe core issue at stake in the discussion about whether
concepts are embodied has been resolved: concepts are
represented in an amodal format^(p. 424).
Similar to Leshinskaya and Caramazza (2016), Mahon
adopts the interaction hypothesis, referring to it as grounding
by interaction (also see Mahon & Caramazza, 2008). Again,
conceptual knowledge and sensory-motor processes reside in
different brain areas and interact when necessary, with inter-
actions between them playing central roles in conceptual pro-
cessing. Nevertheless, Mahon (2015), like Machery (2016),
believes that conceptual processing can function in a stand-
alone manner, stating that,
BThe human mind must have something like a clutch:
something that allows thinking to proceed unencum-
bered by our representations of our body and the world^
Going a step further in this line of argument, Mahon (2015)
presents an account of temporal dynamics called the default
explanation, whereby conceptual processing always occurs first
in conceptual areas and then may later spread optionally to
sensory-motor areas as a function of context (pp. 422, 426-427).
Mahon (2015) further claims that this is the default view of con-
ceptual processing, given that it has been widely established and
accepted (even though important alternative views have existed
for decades; see the discussion of core-first, core-last, and no-core
theories in Lebois et al. 2015; also see Casasanto & Lupyan,
2015; Connell & Lynott, 2014; Yee & Thompson-Schill, 2016).
Finally, Mahon (2015) concludes that researchers must first
reject the default explanation before attempting to show that
alternative grounded theories have merits—the Bburden of
proof^is on these other theories (p. 424). Implicit in this argu-
ment is the assumption that no further evidence is required to
accept the default explanation. The only empirical question is
whether new evidence can be found that rejects it and/or that
supports alternative accounts. As we will see, this implicit as-
sumption has given amodal theories an unfair advantage when it
comes to evaluating evidence and arguments for various theories.
In the process ofmaking these claims, Mahon forces a wide
variety of views onto his Procrustean bed. According to him,
for example, the theories of Barsalou (1999) and Simmons
and Barsalou (2003) contain amodal symbols in convergence
zones. As described later, these theories actually propose that
patterns of conjunctive neurons in convergence zones inte-
grate modality-specific content in a topographical manner,
thereby implementing multimodal compression, not amodal
symbols. Similarly, Mahon argues that Martin (2007,2009)
endorses the interaction hypothesis. As Martin’s(2016)con-
tribution to the special issue clearly indicates, he believes in-
stead that the brain areas active during conceptual processing
are essential parts of modality-specific processing streams and
do not perform stand-alone conceptual processing:
Bmy position is, and has always been, that the regions
where we store information about specific object-
associated properties are located within (i.e., overlap
with) perceptual and action systems, specifically exclud-
ing primary sensory–motor regions.^[italics added]
By forcing these diverse views into the default explanation,
Mahon is contributing to the collection of caricatures and dis-
tortions that the amodal position has produced about grounded
In recent years, Martin (personal communication) has been
asking researchers who use the term Bamodal^what they
mean by it. Overwhelmingly, he finds that they mean
multimodal, not amodal. Sloppy use of Bamodal^has resulted
in this confusing state-of-affairs. To move forward, it is essen-
tial to distinguish carefully between different forms of abstrac-
tion, with amodal symbols constituting one particular form.
An inherent contradiction exists in proposing that amodal processing
areas reside in modality-specific processing streams. To the extent that the
interaction hypothesis is correct, it makes no sense to call these streams
Bmodality specific,^and new ways of referring to them would be re-
quired. When I discuss amodal areas in modality-specific streams here
and at many later points, I’m simply referring to the possibility that
amodal processing areas reside in what have traditionally been called
modality-specific processing streams.
Psychon Bull Rev
The later section on abstraction reviews three other forms of
abstraction, besides amodal symbols, that contributors to the
special issue suggest: multimodal compression, distilled ab-
straction, and distributed linguistic representation. Lumping
these four significantly different forms of abstraction together
as Bamodal^is unlikely to yield progress in understanding
how the brain implements conceptual processing.
In general, two basic assumptions of amodal approaches
make me skeptical that they offer correct and useful accounts
of human conceptual processing: (1) the arbitrary redescrip-
tion of modality-specific content into amodal symbols, and (2)
using these symbols to perform conceptual processing in a
stand-alone manner (at least on some, if not many, occasions).
I continue to doubt that the brain contains amodal conceptual
representations that are arbitrarily related to modality-specific
representations and to their corresponding referents in the
world (Barsalou, 1999). Like Damasio (1989), I don’t see
evidence of any brain regions that implement these kinds of
representations, and I don’t see any compelling rationale for
why they would. Furthermore, I don’t see a compelling ratio-
nale, much less any evidence, for the proposal that amodal
symbols can operate in a stand-alone manner to perform con-
ceptual processing. For these reasons, I view the amodal ap-
proaches reviewed in this section as likely to be Quixotic dead
ends. Later sections flesh out my skepticism further.
Black holes in conceptual space. An intriguing aspect of the
amodal position is that it never provides concrete descriptions
of what amodal concepts are or how they are supposed to work.
None of the articles offered by proponents in this collection
even point to specific proposals, much less begin to develop
them. Machery (2016), for example, simply defines amodal
concepts as having a format that Bdiffers from the format or
formats of perceptual and motor representations,^a remarkably
uninformative definition. Thus, I have become fond of referring
to amodal concepts as black holes in conceptual space. Amodal
concepts appear to have considerable gravitational pull for
some researchers, yet are so mysterious, dark, and dense as
not to be visible and describable. Additionally, the gravitational
fields surrounding these black holes appear to generate consid-
erable distortion in perception, as we have seen.
Seriously, how can we begin to understand and test amodal
theories if it is not clear what they are? Perhaps these re-
searchers simply assume that Fodor's (1975) Language of
Thought is such an obvious solution that they do not need to
mention it (i.e., the view that cognition results from symbolic
operations on structured amodal representations independent-
ly of the modalities and physical situations)? It is this problem,
more than any other, that makes me believe that the amodal
approach is heading up a Quixotic dead end. At least Reilly et
al. (2016) are honest about the problem,
BForemost, the neurobiological mechanisms by which
hubs perform propositional transformations remain es-
sentially a black box.…We must currently take it on
faith that the language of thought involves a form of
mental calculus that operates over abstract symbols:
we have only the most rudimentary understanding of
how the brain extracts and manipulates symbols.^
As we have seen, the offloading hypothesis, the interaction
hypothesis, and the default explanation all assume that the brain
areas containing amodal symbols can support stand-alone con-
ceptual processing. Given the importance of these brain areas in
amodal theories, it is essential that clear coherent accounts of
how they work are forthcoming, as well as empirical predictions
that follow. The fact that such accounts do not exist constitutes a
serious cause for reservation about this class of theories.
If prediction Xfrom grounded cognition is false, then the
amodal view must be true. As we saw earlier, part of being
associated with the default explanation is having a privileged
status when it comes to evidence. All other things being equal,
the amodal view is most likely true by default, especially
when evidence for grounded cognition falls through. No direct
evidence for the amodal view is necessary. Consider a couple
of examples. If an action verb fails to activate motor areas in
an experiment, grounded cognition must be false, and the
amodal view must be true. Analogously, if a motion verb fails
to activate motion areas, again grounded cognition must be
false and the amodal view true. For articulation of this posi-
tion, see Machery (2016).
Clearly, action and motion verbs do activate areas associated
with action and motion, respectively, at least on some occasions
(for many examples of supporting findings, see Dove, 2016,
Jamrozik et al., 2016, and Kemmerer 2015a).
ever, action and motion verbs do not activate motor and motion
I do not include amodal systems for space, time, and magnitude in the
class of amodal symbols that I’m concerned about. In spatial processing,
for example, it’s possible that a single amodal system represents space
across modalities (although this is far from agreed upon; e.g., Braga,
Wilson, Sharp, Wise, & Leech, 2013; Farah, Wong, Monheit, &
Morrow, 1989). Similarly, a single amodal system could represent mag-
nitude across modalities (again this is far from agreed upon, contrary to
Machery’s(2016) claim that consensus exists; e.g., Van Opstal & Verguts,
2013;Walsh,2003; Yates, Loetscher, & Nicholls, 2012). It’s one thing,
however, for amodal systems to represent information that is common
across modalities, and another for amodal symbols to represent modality-
specific information. It is only the latter that I do not believe reside in the
brain. Later, in discussing distilled abstraction, I suggest that the brain
represents many abstract features, similar to space and magnitude,
that do not take the form of transduced amodal symbols (Barsalou,
Although Kemmerer (2015a) is part of the special issue on
Representation of Concepts, it was published earlier in another issue of
Psychonomic Bulletin & Review.
Psychon Bull Rev
areas automatically on all occasions (Tomasino & Rumiati, 2013).
This well-established finding has led some researchers to draw the
default conclusion that if grounded representations are not always
active, then amodal concepts must constitute the conceptual cores
of concepts that do become automatically active across contexts
(Dove, 2009; Machery, 2007; Mahon & Caramazza, 2008).
An emerging theme across multiple literatures, however, is
that concepts do not have conceptual cores, and that even the
most salient features of concepts are not activated automatically
(Gawronski & Cesario, 2013; Kiefer, Adams, & Zovko, 2012;
Lebois et al., 2015; Santiago, Román, & Ouellet, 2011;alsosee
Moors & De Houwer, 2006). Not only is this true of grounded
features, such as action and visual motion, it appears true of all
features, including color features in the Stroop task, positional
features in Simon tasks, numerical features in SNARC tasks,
ward the view that conceptual processing is highly flexible, as
many articles in the special issue illustrate (Binder, 2016;Dove,
2016; Kemmerer, 2015a; Yee & Thompson-Schill, 2016;
Zwaan, 2016; also see Mahon, 2015). Rather than being active
in every context, even the most central features of a concept vary
in how active they are across contexts. Most importantly, it
doesn’t follow from the varying activity levels of grounded fea-
tures that grounded theories are false (Lebois et al., 2015).
Dynamic theories that include grounding naturally explain the
finding that the activation of grounded features varies across
contexts. Even more importantly, it doesn’t follow by default
that amodal theories are correct given these results, because no
direct evidence for amodal symbols has been offered.
If double dissociations in lesion patients occur, then the
amodal view must be true. Another instance of the default
logic concerns double dissociations in lesion patients. First, find
patients who have intact sensory-motor processing but have
conceptual deficits related to these sensory motor processes
(e.g., intact motor processing with deficits in action concepts).
Second, find patients who have sensory-motor deficits but
intact conceptual processing (deficits in motor processing but
intact action concepts). Should these two kinds of patients exist,
then grounded views must be false, and amodal views must be
true. In the special issue, this logic is endorsed for action con-
cepts (Machery, 2016; Leshinskaya & Caramazza, 2016;
Reilly, Peelle, Garcia, & Crutch, 2016; also see Mahon, 2015)
and for food concepts (Rumiati & Foroni, 2016).
There are two problems with this logic. First, it adopts the
dead end perspective on grounded cognition that conceptual
processing is simply equated with sensory-motor processing.
Given this perspective, take away a sensory-motor process,
and any concept that depends on it must no longer exist.
Second, this logic again assumes that if a prediction for the
grounded view fails, the amodal view must be true. It’snot
necessary to actually show that amodal symbols represent
concepts, because this is the default explanation.
Several contributors to the special issue point out that once
plausible grounded views are adopted, it becomes easy to
explain double-dissociations (Kemmerer, 2015a;Martin,2016;
Yee & Thompson-Schill, 2016): When lesion patients have a
deficit on one sensory-motor modality, they use another partially
redundant modality to perform a related conceptual task. When
processing the conceptual meaning of motor verbs, for example,
patients with motor deficits might comprehend motor verbs
using visual motion information. Similarly, when processing
the meanings of food concepts, patients with odor deficits might
comprehend food concepts using visual and taste information.
Another possible explanation follows from adopting an archi-
tecture that contains hierarchically-organized convergence zones
grounded in primary feature areas, with some convergence
zones being modality-specific and others being category-
specific (Simmons & Barsalou, 2003). Depending on where a
lesion falls, numerous combinations of deficits and spared abil-
ities can occur, including the double dissociations just described.
Still another explanation follows from the distributed linguistic
representations described later, with linguistic representations
flexibly supporting conceptual processing in a compensatory
manner when a related modality is damaged.
Although these accounts have been in the literature for years,
and even though multiple contributors to the special issue point
to such accounts, the double dissociation argument lives on. Not
only do double dissociations fail to reject plausible grounded
views, the default explanation simply doesn’t follow without
clear and compelling empirical evidence for amodal symbols.
Having said all this, grounded views generally predict that
significant damage to a modality should have consequences for
conceptual processing. If conceptual processing relies on a mo-
dality—as argued later for neural reuse—then it should change in
some way after the modality is damaged. To assess such conse-
quences, however, it is first necessary to rule out uninteresting
and uncontrolled confounding factors. If, for example, a double
dissociation doesn’t control for the compensatory use of other
modalities, association areas, and/or distributed linguistic repre-
sentations, it is impossible to assess the consequences of a dam-
aged modality. Furthermore, assessments of the damaged modal-
ity should be sufficiently sensitive and well-designed to establish
these consequences. Without designing double dissociation re-
search around these basic principles, it is impossible to assess the
theoretical implications of damage to a modality rigorously.
The fallacy of using conjunction analysis with multimodal
cues to establish amodal symbols in the brain. Another
questionable attempt to support amodal views comes from a
strategy pursued in recent neuroimaging experiments
(Devereux, Clarke, Marouchos, & Tyler, 2013; Fairhall &
Caramazza, 2013; Van Doren, Dupont, De Grauwe, Peeters,
& Vandenberghe, 2010). The basic idea is as follows. First,
identify different cues for a concept that differ in modality-
specific ways (e.g., the spoken word for a concept such as
Psychon Bull Rev
Bdog^vs. a picture of a specific dog instance). Second, present
each cue to participants and establish the parts of the brain that
carry information about the associated concept from that cue.
Third, use conjunction analysis to establish brain areas that
carry information about the concept for both cues. Fourth,
conclude that these shared areas contain amodal symbols.
Thus, the crux of this approach is that any brain area carrying
information about a concept when activated by cues
from different modalities must represent the concept amodally
(yet another case of the default explanation). Machery (2016)
argues that this kind of evidence demonstrates the clear exis-
tence of amodal symbols. Reilly et al. (2016)offeraspecific
example, arguing that brain areas active for both visual and
auditory cues implement amodal symbols.
The obvious problem with this logic is its failure to recog-
nize that different cues could all activate the same modality-
specific information, not just shared amodal symbols. The word
Bdog^and a picture of a dog, for example, could both activate
visual motion areas that represent motion properties associated
with dogs in a grounded manner. If so, then the conjunction
analysis has identified shared modality-specific information,
not the concept’s amodal core. Because researchers performing
these experiments typically do not entertain this possibility, they
don’t attempt to establish whether shared areas in conjunction
analyses process modality-specific or amodal representations.
Thus, again, we do not have compelling evidence for amodal
symbols in the brain, further leading to the conclusion that this
class of theories is heading down a Quixotic dead end.
Promising Approaches to the Study of Concepts
My pessimism about some approaches represented in the spe-
cial issue is more than tempered by my optimism about others.
The large majority of contributions point to future directions
that are likely to be productive, leading to greater understand-
ing of concepts, specifically, and of cognition, in general (in-
cluding contributions of which I’ve been critical so far). I turn
next to themes raised in these contributions that, in my opin-
ion, are likely to be useful elements of future research, includ-
ing grounding, abstraction, flexibility, temporal dynamics, the
ability to explain classic conceptual phenomena, and making
contact with real-world situations.
Many of the contributions in the special issue acknowledge
the importance of grounding conceptual processes and further
acknowledge the considerable evidence for grounding that has
accumulated (Binder, 2016; Dove, 2016; Hauk, 2016;
Jamrozik et al., 2016; Kemmerer, 2015a;Martin,2016;
Reilly et al., 2016; Yee & Thompson-Schill, 2016; Zwaan,
2016). Because some contributions raise issues about what
is meant by grounding (Leshinskaya & Caramazza, 2016;
Martin, 2016;Mahon,2015), I begin with what I mean by it.
To a large extent, grounding concerns itself with the ground-
ing problem raised initially by Searle (1980)andHarnad(1990),
which asks how amodal symbols, specifically, and cognition,
more generally, are linked to the modalities, body, and
environment. In a review of research on grounding, Barsalou
(2008a) argued that researchers have attempted to ground con-
cepts and cognition by establishing their relations with modality-
specific systems, the body, the physical environment, and the
social environment (also see Barsalou, 2010;Barsalouetal.,
2007; Kiefer & Barsalou, 2013). Thus, at a general level,
grounding simply refers to programmatically studying cognition
in new ways. Rather than studying cognitive mechanisms in
isolation, establish their relations with the contexts in which they
are embedded and on which they depend. At more specific
levels, grounding refers to establishing specific accounts of
how cognitive processes in the brain utilize the modalities, the
body, and the environment. It does not mean reducing concepts
and cognition to anything, including sensory-motor mecha-
nisms. As described next, one important form of grounding is
neural reuse, where cognition partially utilizes the modalities to
implement its basic functions (but does not reduce to them).
Neural reuse and simulation. As we saw earlier, the
offloading and interaction hypotheses assume that separate
(and possibly adjacent) brain areas are used for conceptual
and modality-specific processing, such that conceptual pro-
cessing is not grounded—it is self-sufficient, not depending
on any other systems (Leshinskaya & Caramazza, 2016;
Machery, 2016;Mahon,2015). According to this perspective,
processing in conceptual areas doesn’t depend in any way on
modality-specific processing but is simply linked to it.
An alternative view is that conceptual processing depends
on modality-specific systems, utilizing the same—not differ-
ent—systems that support perception, action, and internal
states, what Anderson (2010)calledneural reuse. As Martin
(2016) describes for the domains of color and taste, conceptual
processing in these domains partially utilizes the same sys-
tems used for visually perceiving color and for gustatorally
perceiving taste (see Wang et al., 2013,for another thoughtful
discussion of this issue). Obviously, the neural systems in
conceptual processing and perception differ, given that the
brain is implementing two different processes. Conceptual
processing, for example, is more likely to draw on integrative
and abstractive mechanisms in association areas (Binder,
2016; Simmons & Barsalou, 2003). As a consequence, con-
ceptual processing does not reduce to modality-specific pro-
cessing but utilizes many other systems as well. Nevertheless,
depending on task conditions, conceptual processing, in part,
often reuses systems that underlie perception, action, and in-
Psychon Bull Rev
Neural reuse differs significantly from the idea that amodal
symbols describe modality-specific information in stand-
alone conceptual processing areas (e.g., amodal symbols that
describe color and taste features; Leshinskaya & Caramazza,
2016;Mahon,2015). Instead, neural reuse makes the stron-
ger commitment that modality-specific information is rep-
resented conceptually by partially reusing the same brain
areas that represent this information during perception and
action. Thus, representing color and taste features concep-
tually requires reusing some of the same systems active
during vision and eating.
Neural reuse offers a natural account of what is meant by
simulation (Barsalou, 1999,2003b,2009): Reusing a modal-
ity-specific pathway during conceptual processing simulates
the kind of processing that this pathway performs during per-
ception, action, and/or internal states. Again, reuse may not be
complete and may vary considerably across tasks and con-
texts. Nevertheless, the point remains that higher cognitive
processes are grounded in more basic processes by virtue of
reusing their neural resources. From this perspective, cogni-
tion cannot be performed by only using independent amodal
symbols. Instead, it depends on modality-specific processes to
some extent (along with other mechanisms such compressed
multimodal representations, distilled abstractions, and distrib-
uted linguistic representations, as discussed later).
To the extent that neural reuse plays central roles in con-
ceptual processing, much work must be performed to better
understand it. Specifically, how are conceptual uses of
modality-specific pathways related to uses of these pathways
during perception and action? Are common areas used in sim-
ilar manners, or in different manners? What computational
functions do these areas perform, and how are they related
to modality-specific and conceptual processing? What are
the roles of different laminar layers for bottom-up and top-
down input? In general, we need more sophisticated theoret-
ical accounts of neural reuse that go beyond the simple pro-
posal that neural reuse occurs. Analogously, we need to go
beyond simple demonstration experiments and develop more
systematic empirical investigations that test and inform these
Revisiting the interaction hypothesis. When the first neuro-
imaging results on object concepts were reported, many re-
searchers were surprised that they implicated modality-
specific regions for vision and action (for reviews of such
findings, see Martin, 2007,2016; Martin & Chao, 2001).
Many researchers also were surprised when behavioral re-
search similarly implicated these regions in language compre-
hension (for reviews, see Zwaan, 2004,2016;Zwaan&
Madden, 2005). As this surprise illustrates, researchers ex-
pected to find representations of conceptual knowledge out-
side the brain areas that support modality-specific processing.
It would not have been surprising, for example, if the neural
activity underlying conceptual processing primarily resided in
the higher-level association areas that Binder (2016)reviews,
or in the anterior temporal lobes central to hub-and-spoke
theories (Reilly et al., 2016). Furthermore, because one of
the primary sources of brain expansion has been in association
areas—not in the modalities—it might be expected that asso-
ciation areas would implement the relatively powerful concep-
tual abilities of humans, evolutionarily speaking (Buckner &
For all of these reasons, arguing that brain regions with
amodal symbols reside in what have traditionally viewed as
modality-specific processing streams is unusual and counter-
intuitive. Yet, as we saw earlier, this is the claim of the inter-
action hypothesis (Leshinskaya & Caramazza, 2016;Mahon,
2015). According to this account, amodal regions in modality-
specific processing streams can perform stand-alone concep-
tual processing, only interacting with adjacent modality-
specific areas as needed.
In defending this claim, Leshinskaya and Caramazza
(2016)andMahon(2015) state that grounded researchers
are not justified in making claims about the representational
format used in regions of modality-specific streams active
during conceptual processing. For example, Leshinskaya
and Caramazza (2016)state,
BWe do not see what neural organizing
principles—either the nature of divisions or their
location—can tell us about the nature of those
It follows that grounded researchers are not justified in
claiming that activations for conceptual processing in these
streams are modality-specific. Because the formats of the rep-
resentations in these streams cannot be established, these rep-
resentations could potentially be amodal instead. Notably,
however, the focus of grounded researchers has typically been
on the neural reuse of these regions for conceptual processing,
not on their representational format. Barsalou (1999,p.582),
for example, dismissed the importance of format and stressed
the centrality of neural reuse,
BPerceptual symbols are not like physical pictures; nor
are they mental images or any other form of conscious
subjective experience. As natural and traditional as it is
to think of perceptual symbols in these ways, this is not
the form they take here. Instead, they are records of the
neural states that underlie perception. During percep-
tion, systems of neurons in sensory-motor regions of
the brain capture information about perceived events
in the environment and in the body.^
From this perspective, capturing records of neural states
allows later reusing them for a wide variety of cognitive
Psychon Bull Rev
purposes, without making any commitment to their format
(for discussion of reusing captured neural states,
see Barsalou, 1999 pp. 603-608, 2008a,2009).
Ironically, Leshinskaya and Caramazza (2016)and
Mahon (2015) don’t follow their own admonitions about
the inability to specify format, proposing that conceptual
processing areas in modality-specific streams have an
amodal format (or a non-sensory-motor format). Again,
the default explanation has its privileges. Regardless, these
proposed amodal regions continue to have the character of
black holes in conceptual space. No accounts are provided
of why these regions contain representations in an amodal
format, much less hypothesis-driven empirical evidence for
these claims. Instead, complex elaborate explanations
based on questionable assumptions are used to justify them.
Martin (2016) similarly argues that we can’t determine the
representational format of specific brain regions in modality-
specific processing streams. Nevertheless, he states,
BRepresentations are grounded by virtue of their being
situated within (i.e., partially overlapping with) the neu-
ral system that supports perceiving and interacting with
our external and internal environments.^
Consistent with neural reuse, conceptual processing uti-
lizes modality-specific resources, such that conceptual pro-
cessing overlaps with modality-specific processing. It is not
the format that matters but the fact that conceptual processing
utilizes modality-specific resources—whatever their formats
happen to be. Conceptual processing does not operate in a
stand-alone manner but instead relies on regions of
modality-specific pathways. Martin (2016), together with
many others in the special issue issue and elsewhere, review
considerable evidence consistent with the reuse perspective
(for some recent examples, see Schwiedrzik, Bernstein, &
Melloni, 2016; Waldhauser, Braun, & Hanslmayr, 2016).
One final point about format: Regions of a modality-
specific pathway, such as the ventral stream, have probably
evolved to perform specific types of computations on relevant
information. As a consequence, the representations and pro-
cesses in these regions are likely to reflect such constraints. As
a further consequence, representations and processes are like-
ly to differ across modalities (e.g., vision vs. audition vs. ac-
tion vs. affect vs. taste). If so, then representations and pro-
cesses within a particular modality-specific processing stream
are modality-specific, not amodal. The fact that
cytoarchitectonics vary within and across modalities is consis-
tent with this proposal (Amunts & Zilles, 2015), as is the fact
that different conceptual information can be represented in
different cytoarchitectonic areas (Grill-Spector & Weiner,
2014). Thus, if the neural reuse hypothesis is correct, it fol-
lows that when a conceptual process utilizes the resources of a
modality-specific processing stream, the resultant conceptual
representations have a modality-specific character, not an
Bringing cytoarchitectonics into the discussion may make
it possible to address the format issue in a much more sophis-
ticated way than simply being concerned with whether a brain
area contains amodal or iconic (sensory-motor) representa-
tions. The key issues are likely to be, first, establishing the
neural computations that the cytoarchitectonics of an area
make possible, and second, establishing the cognitive func-
tions these computations realize. I suspect that once we un-
derstand these computations, they will have little to do with
the issue of amodal vs. iconic format. Instead, understanding
these computations is likely to play central roles in under-
standing how conceptual processing reuses modality-specific
Grounding and cognitive processes. Goldinger et al. (2016)
refer to the Bpoverty of embodied cognition.^As Mahon's
(2015) Figure 1 illustrates, articles on embodied cognition
have been growing exponentially since 1980 (this count ap-
parently doesn’t include articles on grounded and situated
cognition). Clearly, some researchers view the grounded per-
spective as having value, and perhaps for reasons described so
far: It’s important to understand how cognition is related to the
modalities, to the body, and to the environment, especially if
cognition depends on them and doesn’t operate independently.
Many researchers believe that the chances of understanding
cognition will increase if we incorporate grounding.
Alternatively, concern exists that cognitive psychology is
not evolving in important new ways, but has become stuck in
an approach that hasn’t changed much in decades.
Traditionally, cognitive psychologists focus on cognitive pro-
cesses per se, ignoring how they’re grounded (e.g., the classic
processes of information processing, such as attention, work-
ing memory, lexical access, etc.). Studying these processes
typically occurs in highly idealized laboratory paradigms de-
signed for the sake of establishing the properties of a particular
process, not because these paradigms reflect important behav-
iors in the world that we need to understand better. When
modeling findings from these paradigms,the resultant theories
typically have little relevance for explaining how the cognitive
processes of interest manifest themselves in real-world behav-
ior. Thus, we end up with theories of laboratory paradigms
whose relations to the world are tenuous. As a further conse-
quence, we accumulate a long list of processes, like those
reviewed in Goldinger et al.’s article, with little understanding
of how they relate to anything, including the modalities, the
body, and the environment. One might worry that this poverty
of riches is leading the field down a Quixotic dead end.
Grounding offers a natural direction for developing ac-
counts of basic cognitive processes. Again, grounded cogni-
tion primarily aims to complement existing accounts, not to
replace them. Not only does grounded cognition benefit from
Psychon Bull Rev
incorporating basic cognitive processes, it offers new oppor-
tunities for studying them, and for understanding how they
operate when embedded in the modalities, body, and environ-
ment. As Barsalou (in press-a) suggests, cognition can be
defined as the mediating processes between stimuli and re-
sponses that make adaptive action in goal-directed situations
possible. As Barsalou (in press-c) further suggests, cognition
is organized around a situation processing architecture that
underlies a wide variety of cognitive, social, affective, and
appetitive behaviors, with neural reuse and simulation playing
central roles. Examining basic cognitive processes from
grounded perspectives like these offers new opportunities for
understanding, studying, and explaining their underlying
mechanisms and their roles in human intelligence.
Perhaps new insights into basic cognitive processes will de-
velop, despite Goldinger et al.’s skepticism. Several of the topics
Goldinger et al. cover—priming, face processing, and sentence
processing—have received considerable attention in grounded
cognition, with researchers believing that something new has
been learned (Lebois et al., 2015; Niedenthal, Mermillod,
Maringer, & Hess, 2010; Santiago et al., 2011; Zwan, 2016).
As researchers have become increasingly convinced that con-
cepts are grounded, they have simultaneously become increas-
ingly aware of how extensively abstraction is associated with
conceptual processing. Indeed, abstraction appears to be a
hallmark of human cognition and an important source of its
computational power. Thus, a current challenge is explaining
how grounding and abstraction emerge together.
It’s perhaps first worth noting that abstraction takes many
forms. Barsalou (2003a), for example, listed six forms of ab-
straction often associated with conceptual processing: (1) cat-
egorical knowledge, (2) the ability to generalize, (3) summary
representation, (4) schematic representation, (5) flexible rep-
resentation, and (6) abstract concepts. Dove (2016)similarly
addresses various forms of abstraction. Contributors to the
special issue bring up all these forms at one point or another
across articles (Binder, 2016; Dove, 2016; Jamrozik et al.,
2016; Leshinskaya & Caramazza, 2016;Martin,2016;
Murphy, 2016; Reilly et al., 2016; Yee & Thompson-Schill,
2016;Zwaan,2016; also see Mahon, 2015).
More significantly, these contributors suggest several
mechanisms that could potentially implement abstraction,
which differ in important ways from amodal symbols: multi-
modal compression, distilled abstraction, and distributed
linguistic representation. Although Mahon (2015)reduces
these different mechanisms into a single construct of amodal
symbols, this move does not offer a useful way forward.
Rather than seeing these mechanisms as the same, it is impor-
tant to understand the differences between them and to
establish their consequences for explaining results, motivating
future research, and for informing our understandings of the
brain and cognition. Certainly, amodal symbols offer one way
of implementing abstraction. Nevertheless, multimodal com-
pression, distilled abstraction, and distributed linguistic repre-
sentation offer important alternatives that appear easier to de-
fine theoretically and that enjoy stronger support empirically.
They also all appear highly compatible with grounded ap-
proaches, as we will see.
Multimodal compression in convergence zones. One ap-
proach to abstracting over the multimodal instances of a cat-
egory is to compress their detailed content into a more general
representation. In the special issue, Binder (2016)proposes
that the brain contains a hierarchical organization of conver-
gence zones, also frequently referred to as association areas
and hubs (e.g., Buckner & Krienen, 2013;Sporns,2010). Of
primary interest for Binder’s account are cross-modal conver-
gence zones whose conjunctive neurons integrate lower-order
information across modalities (also see Damasio, 1989;
Meyer & Damasio, 2009; Simmons & Barsalou, 2003).
Within these convergence zones, patterns of conjunctive neu-
rons establish cross-modal conjunctive representations
(CCRs) that function as abstract representations. In the limit,
Binder assumes that CCRs can become so abstract as to some-
times become amodal symbols. More typically, however, he
assumes that they maintain information about the lower-order
representations they integrate across modalities, thereby
exhibiting data compression rather than arbitrary transduction
(Barsalou, 1999). Rather than being arbitrarily related to multi-
modal simulations, as for amodal symbols, CCRs evolve out of
multimodal simulations, continuing to carry information about
them in a representational manner (perhaps this is the crux of
what is meant by a convergence zone). In this spirit, Simmons
and Barsalou (2003) proposed that conjunctive neurons are
organized topographically within convergence zones, where to-
pographical closeness can be defined by conceptual similarity as
well as by modality-specific similarity. Binder further suggests
that CCRs support the computations underlying categorical
representations, thematic relations, propositions, and schemata.
Binder reviews much literature that bears on the localiza-
tion of multimodal compressions, such as CCRs, in the brain.
Specifically, this literature suggests that the associative pro-
cesses underlying conceptual knowledge consistently engage
brain areas outside modality-specific areas, including medial
prefrontal cortex (dorsal and ventral), posterior cingulate cor-
tex, angular gyrus/supramarginal gyrus, lateral middle tempo-
ral gyrus, inferior frontal gyrus, and ventromedial temporal
cortex (also see Legrand & Ruby, 2009). In general, these
areas tend to become more active for stimuli associated with
many abstractions than for stimuli associated with fewer ab-
stractions. Specifically, these areas tend to become more ac-
tive for meaningful words than for pseudo-words (and for
Psychon Bull Rev
meaningful sentences and texts than for anomalous ones). The
same areas are more active for familiar proper names and
high-frequency words than for unfamiliar proper names and
low-frequency words, and when highly related and associated
concepts are processed together than when less related and
associated concepts are processed together. Finally, process-
ing concepts deeply activates these areas more than when the
same concepts are processed shallowly.
Based on these findings, Binder concludes that these asso-
ciation areas process abstractions. As the abstractions associ-
ated with a stimulus increase, these areas become more active,
because they represent and process the increasing numbers of
abstractions. Additional findings from Fernandino et al. (in
press) further implicate compressed multimodal representa-
tions in association areas, showing that association areas all
over the brain track the modality-specific content of concepts.
Rather than being arbitrarily related to modality-specific con-
tent in concepts, association areas represent it.
Compressed representations, such as CCRs, are essentially
the same kind of representations as prototypes in cognitive
theories of concepts (Murphy, 2016). According to prototype
theory, statistically likely features are extracted from category
exemplars and conjoined in a prototype that represents the
category conceptually (Hampton, 2006; McRae, Cree,
Seidenberg, & McNorgan, 2005;Rosch&Mervis,1975;
Smith & Medin, 1981). Notably, prototypes are not amodal
symbols arbitrarily linked to examplars. Instead, the features
of exemplars appear in the prototype that covers exemplars,
following various possible forms of data compression, as for
An alternative form of data compression that could underlie
multimodal abstractions reduces exemplar information to a
new set of dimensions (as in PCA, ICA, NMF, and so forth).
Rather than using features from exemplars to represent a pro-
totype, new dimensions are abstracted that capture correlated
sets of features. As a result, different vectors across these new
dimensions would represent concepts in Binder’sassociation
areas. Again, these compressions aren’tamodalsymbols,be-
cause they carry information about the information in exem-
plars. For examples of how this type of compression can be
used to represent different kinds of concepts, see Reilly et al.
(2016), Blouw et al. (2015), and Mitchell et al. (2008).Many
neural net models similarly use dimensional reduction to repre-
sent knowledge (Ganguli & Sompolinsky, 2012; Hinton, 2006).
As Murphy (2016) notes, exemplar theories offer an alter-
native to prototype theories, with exemplar theories assuming
that individual exemplar memories represent a category in-
stead of an aggregate prototype. Although Murphy has strong
inclinations to view exemplar theories as a Quixotic dead end,
the mechanisms that integrate exemplars into a category rep-
resentation could potentially reside in the areas that Binder
(2016) associates with abstraction, if these theories turned
out to be accurate.
Murphy further suggests that exemplar effects offer a plau-
sible alternative to exemplar theories, with salient exemplars
sometimes controlling categorization (Allen & Brooks, 1991).
A class of accounts that naturally explains both prototype and
exemplar effects assumes that abstractions include statistical
information about correlations of features, not just about
individual features (Barsalou, 1990;Barsalou&Hale,1993;
McClelland & Rumelhart, 1985; also see Love, Medin, &
Gureckis, 2004). Because exemplars can be viewed as pat-
terns of correlated features, storing correlational information
from them in abstractions implements exemplar effects (e.g.,
using hidden units; Gotts, 2016). Factorization approaches,
such as PCA, ICA, and NMF, also are readily capable of
capturing exemplar effects, representing salient exemplars
with individual components. Again, such patterns constitute
data compression that could plausibly be implemented in
Binder’s(2016) association areas. As all these approaches
indicate, it is possible to capture the exemplar effects that
Murphy endorses, without further endorsing exemplar models
that posit individual exemplar representations in memory.
Distilled abstractions. In their account of metaphor, Jamrozik
et al. (2016) suggest another mechanism for abstraction that
differs from multimodal compression. Rather than
compressing multimodal information into a sparser represen-
tation, abstraction distills abstract features from more concrete
ones, leaving behind a representation that contains only ab-
stract features. Although Jamrozik et al. focus on abstract
features transmitted into a concept via metaphorical compari-
son, in principle, any concept that contains both concrete and
abstract features could have its concrete features filtered, leav-
ing an abstract representation behind (cf. Barsalou, 1999,pp.
What sorts of abstract features exist that could be distilled
into abstractions? As discussed earlier, recent research has
already established that the brain represents a variety of ab-
stract features associated with the concepts of convince, arith-
metic, goal, belief,andfunction (Leshinskaya & Caramazza,
2016; Wilson-Mendenhall et al., 2013). The abstract features
of reward and magnitude offer further examples of important
abstract features, given that dedicated brain areas represent
them to some extent, namely, the orbital frontal cortex and
intraparietal sulcus, respectively (Rudebeck & Murray, 2014;
Wals h , 2003; Wilson, Takahashi, Schoenbaum, & Niv, 2014).
Potentially, many diverse abstract features structure experi-
ence extensively and thus play central roles in distilled
Linguistic analyses of syntactic and semantic primitives
offer another possible source for discovering abstract features.
Talmy (1985), for example, offered numerous examples of
abstract features that are lexicalized across languages, playing
a wide variety of important roles in syntax and semantics,
including features for cause, aspect, valence, together with
Psychon Bull Rev
more concrete features such as path, motion, and manner (also
see Langacker, 1986,2008, and many other cognitive
linguists). Increasing research establishes the neural systems
that implement these features (Kemmerer, 2006;Kemmerer,
Castillo, Talavage, Patterson, & Wiley, 2008).
What is the status of abstract features in the brain? Because
these features represent abstract information present in a situ-
ation, perhaps they are similar in status to sensory-motor fea-
tures, such as color, motion, pitch, taste, touch, and movement
that represent features in perception and action. As instances
of these features are experienced, the corresponding feature
areas become active to represent them. Alternatively, amodal
symbols could represent abstract features, a common assump-
tion that many researchers make. Nevertheless, it seems im-
portant and potentially useful to consider the possibility that
abstract features have a similar status in the brain as sensory-
motor features, coding aspects of experience directly, rather than
being represented by amodal symbols that have been somehow
transduced from experience (Barsalou, 1999;Barsalou&
Wiemer-Hastings, 2005; Wilson-Mendenhall et al., 2013).
On subsequent occasions, when abstract features of entities
and events are situations are represented conceptually, the
areas associated with processing abstract features are
reactivated, simulating what those situations were like ab-
stractly. Similar to sensory-motor features, neural reuse sup-
ports the conceptual processing of abstract features. Many of
these features are likely to have strong biological origins (e.g.,
magnitude, reward, mental state).
A key issue is whether abstract features reside in the asso-
ciation areas that Binder (2016) reviews or elsewhere.
Preliminary evidence suggests that they can reside in both.
Consider findings from Wilson-Mendenhall et al. (2013).
For the abstract concept, arithmetic, they found activations
in the intraparietal sulcus that overlapped with activations
for a numerical localizer task. Because this brain area does
not fall within Binder’s association areas, and because it is
often associated with the representation of magnitude, it fol-
lows that some abstract features are processed by dedicated
feature areas. Again, it seems important to enumerate these
kinds of features, to establish where they reside in the brain,
and how they contribute to abstractions.
Wilson-Mendenhall et al., however, observed a different
finding for the activations associated with the abstract con-
cept, convince. In general, activations for these areas fell
largelyintoBinder’s association areas (with some excep-
tions). Because these activations also overlapped with ac-
tivations for a mental states localizer task, it is not clear
whether the activations for convince reflect data compres-
sion versus distilled abstraction for features of mental
states. Thus, another important issue for future research is
to establish whether Binder’s association areas (and asso-
ciation areas more generally) perform data compression,
represent abstract features, or both.
Distributed linguistic representations. In the special issue,
Zwaan (2016) suggests that distributed linguistic representa-
tions can function as symbolic placeholders, which become
fleshed out by multimodal simulations when greater informa-
tion about a concept is needed. Although this account is sim-
ilar to multimodal compression, it assumes that a completely
different kind of representation—associated sets of words—
function as a compressed account of conceptual content rather
than conjunctive patternsin a convergence zone at the end of a
processing stream. Many other researchers have made similar
claims, arguing that distributed linguistic representations play
central roles in conceptual processing (Andrews, Frank, &
Vigliocco, 2014; Andrews, Vigliocco, & Vinson, 2009;
Barsalou et al., 2008; Connell & Lynott, 2013;Glaser,1992;
Louwerse, 2011; Louwerse & Connell, 2011; Paivio, 1986);
also see research on Latent Semantic Analysis and related
approaches (Baroni & Lenci, 2010;Erk,2012; Erk & Padó,
Dennis, & Kintsch, 2013; Padó & Lapata, 2007). Because
the words that constitute distributed linguistic representations
typically take auditory, motor, and visual forms, they are
modality-specific representations, not amodal symbols.
In general, this approach proposes that computations over
distributed linguistic representations can be used to perform
cognitive tasks effectively, although many researchers suggest
that the effectiveness is less than when using multimodal sim-
ulations and other forms of conceptual knowledge (Barsalou
et al., 2008; Connell & Lynott, 2013; Glaser, 1992; Louwerse
&Connell,2011; Zwaan, 2016). In other words, distributed
linguistic representations offer a heuristic for performing a
task quickly, when conditions permit, but may not be suffi-
cient for sophisticated task performance. Again, they can be
viewed as abstractions that carry information about more com-
plex conceptual representations.
Establishing the distributed networks that underlie con-
cepts. According to previous sections, a concept utilizes neu-
ral resources that represent features in modality-specific sys-
tems, together with various kinds of abstract features.
Furthermore, these features feed into association areas that
compress incoming information, perhaps at multiple levels,
resulting in compressed abstractions. Abstractions of a con-
cept can also result when abstract features are distilled from
concrete ones, and when distributed linguistic representations
are established for it.
If this account of concepts is correct, then establishing the
distributed networks that implement neural reuse, data com-
pression, distilled abstraction, and distributed linguistic repre-
sentations presents a number of research challenges. One im-
portant challenge is to establish the association areas that play
central roles in abstraction. Because a variety of conflicting
proposals exist, assessing their relative merits is essential to
producing a compelling account of these distributed networks.
Psychon Bull Rev
Hub-and-spoke theories constitute one important class
of theories. According to these theories, the most important
association areas for concepts reside in the anterior tempo-
ral lobes (ATL) of both the left and right hemispheres
(Reilly et al., 2016; also see Lambon Ralph, Sage, Jones,
& Mayberry, 2010; Patterson, Nestor, & Rogers, 2007;
Rogers & McClelland, 2004). From this perspective, the
ATLs contain stand-alone amodal symbols that can per-
form conceptual processing on their own, but that project
to sensory-motor areas when fleshing out amodal symbols
is useful (similar to the interaction hypothesis). Although
hub-and-spoke theories assume that the abstractions in the
ATLs are amodal, the ATLs also could also potentially
implement multimodal data compressions and/or distilled
Perhaps a more serious problem for hub-and-spoke the-
ories than its assumptions about amodal symbols is that
many researchers do not view the ATL as a single homo-
geneous region for representing all concepts in general, but
rather as a highly differentiated region where distinct con-
cepts, such as those for social information and individuals,
are represented separately from other concepts (Drane
et al., 2008,2009,2013; Martin, Simmons, Beauchamp,
&Gotts,2014; Wong & Gallate, 2012). Furthermore, when
Fernandino et al. (in press) used conjunction analysis to
establish common association areas active across diverse
concepts, the ATLs were not present. Binder (2016)suggests
that other association areas in the temporal and parietal lobes
are most likely to perform Bhub^functions (also see Binder,
Desai, Graves, & Conant, 2009). Martin (2016) similarly sug-
gests that other temporal and parietal areas operate as hubs.
Once the association areas for conceptual processing are
established, it then becomes important to establish whether
they support data compression, distilled abstraction, distribut-
ed linguistic representation, or some other function (including
amodal symbols). Once the operative representations in an
associative area are established, another critical issue is deter-
mining whether these representations can operate in a stand-
alone manner or can only operate together with modality-
specific representations. Whenabstractionsandsensory-
motor simulations operate together, the nature of their interac-
tions becomes important, along with how these interactions
accomplish various conceptual functions (similar to issues
associated with the interaction hypothesis).
Following Jamrozik et al. (2016), distilled abstractions can
probably operate as stand-alone representations, given that
they sometimes become active in metaphor without the sup-
port of activations in modality-specific areas (which can ac-
company them optionally). As Zwaan (2016) points out, how-
ever, abstract representations often become much easier to
understand, once they have been embellished with modality-
specific information (also see Akpinar & Berger, 2015;
Distributed linguistic representations may also be capable
of functioning as stand-alone representations, following many
recent empirical demonstrations (Zwaan, 2016;alsosee
Barsalou et al., 2008; Connell & Lynott, 2013; Louwerse &
Jeuniaux, 2010). Again, they may only provide heuristic in-
formation about a concept, with more definitive representation
requiring the addition of multimodal simulations.
Finally, it is not clear that multimodal compressions in as-
sociation areas can function as stand-alone representations
during conceptual processing, or whether they require support
from modality-specific areas. One proposal is that compressed
representations primarily serve to index, integrate, and control
distributed representations across multimodal systems rather
than to represent them in a stand-alone manner (Damasio,
1989; Simmons & Barsalou, 2003). Nevertheless, because
multimodal compressions carry information about distributed
multimodal representations, they could potentially be used in
a stand-alone manner, at least for heuristic purposes.
Temporal dynamics. As Hauk (2016) notes, various distrib-
uted networks just considered make temporal predictions
about neural activity and behavior. Hub-and-spoke theories,
for example, predict that hub representations typically become
active before spoke representations. Similarly, the default ex-
planation proposes that conceptual areas become active before
sensory-motor areas. Finally, research on distributed linguistic
representations has shown that, at least under some condi-
tions, distributed linguistic representations become active be-
fore conceptual representations, such as multimodal
Another possibility—consistent with research on flexibility
discussed shortly—is that the networks underlying concepts
can be accessed in diverse ways, such that no strict time course
characterizes their operation. In this spirit, Zwaan (2016)il-
lustrates how the processing of an abstract concept could dif-
fer when used cataphorically versus anaphorically (i.e., when
an abstract concept is processed before or after a relevant
situation in which it applies). When processed cataphorically,
a sparse representation is accessed initially (perhaps a distrib-
uted linguistic representation), which is then fleshed out in
modality-specific detail with the text that follows.
Conversely, when processed anaphorically, relevant
modality-specific detail is established before the abstract con-
cept is presented, such that a rich conceptual representation is
available immediately. To the extent that such flexibility ex-
ists, taking tasks and contexts into account is essential for
making predictions about temporal dynamics.
Predictions about temporal dynamics offer an important
tool for confirming and disconfirming hypotheses about the
distributed networks that underlie concepts. Localization of
these networks spatially is obviously important but limited
in the information provided. Assessing temporal dynamics
not only offers additional means of discriminating theories,
Psychon Bull Rev
it also forces researchers to articulate their theories more clear-
ly. Developing detailed models of the types that Gotts (2016)
explores is likely to support making sophisticated predictions
about temporal dynamics.
Abstract concepts. Many contributors to the special issue
note the importance of abstract concepts in assessing the-
ories of concepts (Dove, 2016;Jamroziketal.,2016;
Leshinskaya & Carmazza, 2016; Martin, 2016; Reilly
et al., 2016; Zwan, 2016; also see Mahon, 2015). These
contributors often note that abstract concepts are ubiqui-
tous in human cognition, perhaps playing more important
roles than concrete concepts. Even more often, abstract
concepts are presented as a specific challenge for ground-
ed theories, even though they actually constitute a major
challenge for all theories. To my knowledge, the amodal
approach has offered even less insight into abstract con-
cepts at this point than grounded approaches, given the
current literature. Again, the default assumption often is
that if the grounded approach cannot explain something,
then the answer must be amodal. So goes the common
story about abstract concepts. Regardless, the amount that
we understand about abstract concepts from any perspec-
tive is shockingly modest. Simply claiming that abstract
concepts are represented by amodal symbols does not ex-
plain anything about their content, structure, and function.
Abstract concepts seem like a particularly hard nut to
crack. Our lack of understanding doesn’t reflect a lack of
effort but more likely reflects what a hard problem this is.
If we want to understand human concepts—and indeed
human cognition—it is essential that we establish some
traction in understanding abstract concepts. I suspect that
all the abstraction mechanisms described earlier play im-
portant roles. Because abstract concepts often appear to
integrate information across situations, they may utilize
multimodal compression. Because they often include ab-
stract features, they may utilize distilled abstraction.
Because they appear highly dependent on language
(Binder, Westbury, McKiernan, Possing, & Medler,
2005), they may utilize distributed linguistic representa-
tions. Consistent with the multifaceted character of abstract
concepts, both Dove (2016) and Jamrozik et al. (2016)
propose that multiple mechanisms are likely to underlie
Moving beyond our modest understanding of abstract con-
cepts may require studying a few of them in detail, rather than
studying them in the aggregate, especially given their diversi-
ty (Wilson-Mendenhall et al., 2013). Because abstract con-
cepts appear dependent on background situations (Akpinar
&Berger,2015; Schwanenflugel, 1991;Zwaan,2016), exam-
ining their operation in specific situations may provide lever-
age—and perhaps be necessary—for making significant
Another major theme across contributions to the special issue is
that concepts operate flexibly, with their representations and ef-
fects depending on tasks and contexts (Binder, 2016; Dove,
2016; Kemmerer, 2015a; Yee & Thompson-Schill, 2016;
Zwaan, 2016; also see Mahon, 2015). In their review of context
effects on concepts, Yee and Thompson-Schill illustrate that the
processing of a concept varies as a function of long-term context,
recent context, immediate context, ongoing context, and individ-
ual abilities. They further argue that (1) concepts cannot be sep-
arated from the contexts in which they occur (consistent with
context availability theory; Schwanenflugel, 1991), (2) concepts
are not rigidly fixed in content, and (3) concepts do not have
conceptual cores. Yee and Thompson-Schill further suggest that
recurrent neural networks offer a natural way of explaining how
all the different levels of context shape the knowledge underlying
a concept and its changing content over a processing episode.
Kemmerer (2015a) first reviews evidence that motor and
motion areas frequently become active to represent the mean-
ings of action and motion verbs. Then, focusing on action
verbs, he shows that activations in motor areas do not always
occur when processing them. Analogous to Yee and
Thompson-Schill, he reviews a wide variety of factors that
modulate motor activations during meaning activation.
Motor activations only represent verbs, for example, under
conditions when they are relevant and available for doing so.
Otherwise, verb meanings are represented flexibly using other
mechanisms along the lines of those just discussed for abstrac-
tion. After briefly reviewing evidence that color information is
not always active in the Stroop paradigm, Kemmerer states,
BSurely, the lack of a consistent interference effect in the
Stroop paradigm does not imply that the color features
of color words are not genuine components of the mean-
ings. By the same token, the discovery that the motor
features of action verbs (and tool nouns) are not always
accessed in the same way does not imply that those
features are not genuine components of the meanings.^
Based on a review of automaticity in the semantic process-
ing literature, Lebois et al. (2015) similarly argue that concepts
don’t have conceptual cores that are activated immediately
across all contexts. Instead, even the most central features of
a concept can vary widely in activation. They further propose
that the flexible activation of information in a concept reflects
Bayesian sampling: On a given occasion, the probability that a
given feature of a concept is active reflects (1) the overall
frequency with which the feature has been processed in pre-
vious contexts (its prior) and (2) its relevance in the current
context (its likelihood). Although frequent features have a
high chance of becoming active, context can override them,
conferring an advantage on contextually relevant features. As
Psychon Bull Rev
context becomes increasingly salient and specified, con-
textually relevant features may become increasingly dom-
inant, similar to well-known context effects in the Visual
World Paradigm (Tanenhaus, Spivey-Knowlton, Eberhard,
&Sedivy,1995; for a recent review see, Huettig,
Rommers, & Meyer, 2011)Basedonawidevarietyofsuch
findings, Hauk (2016), Yee and Thompson-Schill (2016),
and Lebois et al. (2015) suggest that understanding the
dynamic activation of features for a concept in context over
time should be a central topic of future research (for a
recent example of such work, see van Dam, Brazil,
Bekkering, & Rueschemeyer, 2014).
Finally, Gotts (2016) provides elegant and insightful re-
views of repetition priming in behavior, together with the cor-
responding repetition suppression in neural activity, exploring
the implications for incremental learning. Although the
models that he reviews and develops primarily address the
processing of individual stimuli, the underlying mechanisms
have clear relevance for the representation of conceptual
knowledge, as he notes. In addition, these mechanisms pro-
vide a natural way to understand and model the kinds of con-
ceptual flexibility that Kemmerer (2015a), Yee and
Thompson-Schill (2016), and Lebois et al. (2015) review.
Our understanding of conceptual flexibility will be enhanced
significantly to the extent that we can explain the phenomena
reviewed in these articles with the kinds of models that Gotts
develops. Implementing such models using the network archi-
tectures motivated earlier in the sections on grounding and
abstraction offers one direction for model development.
Because these architectures implement multiple forms of rep-
resentation across association areas and modalities, a given
concept can be flexibly activated in many different ways.
Explaining Classic Conceptual Phenomena
In the special issue, Binder (2016) and Murphy (2016)remind
us that theories of concepts must explain classic conceptual
phenomena (Barsalou, 2003b,2012; Margolis & Laurence,
1999;McRae&Jones,2013;Murphy,2002). As Binder re-
views, concepts play important roles in taxonomies, thematic
relations, propositions, and schemata. Murphy also addresses
the importance of taxonomies, along with basic level catego-
ries, exemplar effects, and category-based induction. In par-
ticular, Murphy emphasizes the importance of causal reason-
ing and background knowledge in conceptual processing.
Typicality and all its effects would be another basic phenom-
enon that I’d add to this list.
I’d further add basic symbolic operations, such as type-
token binding, predication, propositions, productivity, and
concept composition (Fodor, 1975; Fodor & Pylyshyn,
1988;Pylyshyn,1984). Although it might seem that symbolic
operations are only relevant to amodal theories, and can only
be explained by them, I continue to believe that they funda-
mentally structure human cognition, and that grounded theo-
ries have potential to explain them in novel and insightful
ways (Barsalou, 1999,2008b,in press-b). As Binder
(2016) suggests, another important function of association
areas is to integrate conceptual information (also see
Humphries, Binder, Medler, & Liebenthal, 2006;
Legrand & Ruby, 2009).
Although these conceptual phenomena may seem dated,
the only things dated are our theories of them. As current
theories evolve, it’s essential that they continue to offer com-
pelling accounts of foundational phenomena associated with
Making Contact with Real-World Situations
Cognitive psychologists typically assume that the kinds of
phenomena listed in Goldinger et al. (2016) generalize readily
to real-world situations. As a consequence, no attempt is made
to assess these mechanisms in richer, more natural situations
outside the idealized laboratory environments in which they
were established. As Henrich, Heine, and Norenzayan (2010)
found, however, such assumptions are not always warranted,
given that a given cognitive mechanism can take different
forms outside the laboratory.
The Visual World Paradigm offers another good example
of how a basic cognitive phenomenon can change significant-
ly when embedded in more realistic situations (Huettig et al.,
2011; Tanenhaus et al., 1995). Prior to the visual world para-
digm, what we knew about the processing of words and syn-
tax was based on paradigms that largely examined them in
isolation. Once researchers began studying them in the context
of visual worlds, understanding evolved considerably to re-
flect substantial contributions from context.
Because concepts are typically studied in relatively ideal-
ized laboratory tasks, accounts of them may similarly be lim-
ited; we may be missing important phenomena and experienc-
ing a variety of distortions. Although many of the basic con-
ceptual phenomena established so far are likely to remain, it
wouldn’t be surprising if new findings and insights followed
As Rumiati and Foroni (2016) illustrate, food concepts
offer an example of how concepts can be studied in a rich
real-world domain (also see Papies, 2013;Papies&
Barsalou, 2015; Ross & Murphy, 1999). Because eating is a
multifaceted phenomenon that utilizes many cognitive pro-
cesses, including attention, categorization, learning, decision
making, goal pursuit, and affect, it offers the opportunity to
examine how concepts operate in a rich cognitive context that
also happens to be of social significance.
Many other real-world domains similarly offer useful con-
texts for increasing our understanding of how concepts
Psychon Bull Rev
operate in the real world. Barsalou et al. (2007) suggested two
general types of situations that are central in human activity:
(1) A single individual performs situated action in the physical
environment to achieve a goal, as when building an object,
navigating to a location, or gathering resources; (2) Multiple
individuals perform coordinated social interaction, as when
performing knowledge transfer, group decision making, or
distributed operation of complex equipment. Studying how
concepts support intelligent activity in situations like these
might lead to new insights about their content, structure, and
function, which wouldn’t emerge from studying them in the
I’m not suggesting that basic cognitive research become
applied. Instead, I’m suggesting that doing a better job of
linking basic research to the real world is likely to have sig-
nificant benefits for basic research. Bringing cognitive theory
and results to bear on real-world problems will not only pro-
duce social benefits, it will increase the visibility of cognitive
research and perhaps bring it greater appreciation and re-
sources. Furthermore, examining basic cognitive processes
in real world situations offers opportunities to assess whether
our basic accounts are correct and complete, and if not, where
further development is needed. To the extent that an account
of a cognitive process is inadequate, assessing its behavior in
the real world is likely to inform how the account could be
revised, leading to important new basic research opportunities
in the laboratory. In these manners, connecting basic research
to important real world situations has significant potential to
keep basic research from becoming insular, and to promote its
productive growth and evolution.
I began with Quixotic dead ends for studying concepts, name-
ly, principled approaches that are unlikely to succeed. In this
category, I included grounded theories that reduce to sensory-
motor processes, grounded theories without representation or
biological constraints, grounded theories that focus only on
embodiment, and grounded theories that aim to replace stan-
dard accounts of cognition rather than complement them. In
the category of Quixotic dead ends, I also included amodal
theories with arbitrary symbols that can operate in a stand-
alone manner, and approaches to cognitive psychology that
simply develop lists of cognitive processes established in ide-
alized laboratory paradigms unrelated to real-world situations.
Based on contributions to the special issue, I developed an
approach to studying concepts that I believe has some chance
of heading in the correct direction and enjoying some success.
The properties of this approach include grounding, abstrac-
tion, flexibility, the ability to explain classic conceptual phe-
nomena, and making contact with real-world situations.
Having a few Quixotic bones in my body, I realize where I
may be headed. Hopefully, I’m sufficiently grounded so that
whatever my mistakes may be, they’re not too misleading and
are useful for others in finding a clearer path.
Acknowledgments The author thanks Esther Papies and Philippe
Schyns for helpful discussion and comments on an earlier draft, and
Greg Hickok, David Kemmerer, Brad Mahon, Alex Martin, and Greg
Murphy for expert reviews and discussion.
Open Access This article is distributed under the terms of the Creative
Commons Attribution 4.0 International License (http://
creativecommons.org/licenses/by/4.0/), which permits unrestricted use,
distribution, and reproduction in any medium, provided you give appro-
priate credit to the original author(s) and the source, provide a link to the
Creative Commons license, and indicate if changes were made.
Adams, S. V., Wennekers, T., Cangelosi, A., Garagnani, M., &
Pulvermuller, F. (2014). Learning visual-motor cell assemblies for
the iCub robot using a neuroanatomically grounded neural network.
In 2014 I.E. Symposium on Computational Intelligence, Cognitive
Algorithms, Mind, and Brain (CCMB) (pp. 1–8).
Akpinar, E., & Berger, J. (2015). Drivers of cultural success: The case of
sensory metaphors. Journal of Personality and Social Psychology,
Allen, S. W., & Brooks, L. R. (1991). Specializing the operation of an
explicit rule. Journal of Experimental Psychology: General,
Amunts, K., & Zilles, K. (2015). Architectonic mapping of the human
brain beyond Brodmann. Neuron, 88, 1086–1107.
Anderson, M. L. (2010). Neural reuse: A fundamental organizational
principle of the brain. Behavioral and Brain Sciences, 33, 245–266.
Andrews, M., Frank, S., & Vigliocco, G. (2014). Reconciling embodied
and distributional accounts of meaning in language. Topics in
Cognitive Science, 6, 359–370.
Andrews, M., Vigliocco, G., & Vinson, D. (2009). Integrating experien-
tial and distributional data to learn semantic representations.
Psychological Review, 116, 463–498.
Baroni, M., & Lenci, A. (2010). Distributional memory: A general frame-
work for corpus-based semantics. Computational Linguistics, 36,
Barsalou, L. W. (in press-a). Can cognition be reduced to action?
Processes that mediate stimuli and responses make human action
possible. In A. K. Engel, K. J. Friston, & D. kragic, Where’sthe
action? The pragmatic turn in cognitive science (Strüngmann
Forum Reports, Vol. 18. J. Lupp, Series Ed.). Cambridge, MA:
Barsalou, L. W. (in press-b). Cognitively plausible theories of concept
composition. In Y. Winter & J. A. Hampton, Compositionality and
concepts in linguistics and psychology. London: Springer
Barsalou, L. W. (in press-c). Situated conceptualization: Theory and ap-
plications. In Y. Coello & M. H. Fischer, Foundations of embodied
cognition. East Sussex: Psychology Press.
Barsalou, L. W. (1990). On the indistinguishability of exemplar memory
and abstraction in category representation. In T. K. Srull & R. S. W.
Jr, Content and process specificity in the effects of prior experiences:
Advances in social cognition (Vol. 3, pp. 61–88). Hillsdale, NJ:
Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and
Brain Sciences, 22, 577–660.
Psychon Bull Rev
Barsalou, L. W. (2003a). Abstraction in perceptual symbol systems.
Philosophical Transactions of the Royal Society of London. Series
B, Biological Sciences, 358, 1177 –1187 .
Barsalou, L. W. (2003b). Situated simulation in the human conceptual
system. Language and Cognitive Processes, 18, 513–562.
Barsalou, L. W. (2008a). Grounded cognition. Annual Review of
Psychology, 59, 617–645.
Barsalou, L. W. (2008b). Grounding symbolic operations in the
brain’s modal systems. In G. R. Semin & E. R. Smith (Eds.),
Embodied grounding: Social, cognitive, affective, and neurosci-
entific approaches (pp. 9–42). New York: Cambridge University
Barsalou, L. W. (2009). Simulation, situated conceptualization, and pre-
diction. Philosophical Transactions of the Royal Society, B:
Biological Sciences, 364, 1281–1289.
Barsalou, L. W. (2010). Grounded cognition: past, present, and future.
Topics in Cognitive Science, 2, 716–724.
Barsalou, L. W. (2012). The human conceptual system. In M. Spivey, K.
McRae, & M. F. Joanisse (Eds.), The Cambridge handbook of
psycholinguistics (pp. 239–258). New York: Cambridge University
Barsalou, L. W., Breazeal, C., & Smith, L. B. (2007). Cognition as coor-
dinated non-cognition. Cognitive Processing, 8, 79–91.
Barsalou, L. W., & Hale, C. (1993). Components of conceptual represen-
tation. From feature lists to recursive frames. In I. Van Mechelen, J.
A. Hampton, R. Michalski, & P. Theuns (Eds.), Categories and
concepts: Theoretical views and inductive data analysis (Hale, C,
pp. 97–144). San Diego: Academic Press.
Barsalou, L. W., Santos, A., Simmons, W. K., & Wilson, C. D. (2008).
Language and simulation in conceptual processing. In M. De Vega,
A. M. Glenberg, & A. C. Graesser (Eds.), Symbols, embodiment,
and meaning (pp. 245–283). Oxford: Oxford University Press.
Barsalou,L. W., Simmons, W. K., Barbey, A. K., & Wilson, C. D. (2003).
Grounding conceptual knowledge in modality-specific systems.
Trends in Cognitive Sciences, 7, 84–91.
Barsalou, L. W., & Wiemer-Hastings, K. (2005). Situating abstract con-
cepts. In D. Pecher & R. A. Zwaan, (pp. 129–163). New York:
Cambridge University Press.
Binder, J. R. (2016). In defense of abstract conceptual representations.
Psychonomic Bulletin & Review.doi:10.3758/s13423-015-0909-1.
Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where
Is the semantic system? A critical review and meta-analysis of 120
functional neuroimaging studies. Cerebral Cortex, 19, 2767–2796.
Medler, D. A. (2005). Distinct brain systems for processing
concrete and abstract concepts. Journal of Cognitive
Neuroscience, 17, 905–917.
Blouw, P., Solodkin, E., Thagard, P., & Eliasmith, C. (2015). Concepts as
semantic pointers: A framework and computational model.
Cognitive Science, 1–35.
Braga, R. M., Wilson, L. R., Sharp, D. J., Wise, R. J. S., & Leech, R.
(2013). Separable networks for top-down attention to auditory non-
spatial and visuospatial modalities. NeuroImage, 74, 77–86.
Buckner, R. L., & Krienen, F. M. (2013). The evolution of distributed
association networks in the human brain. Trends in Cognitive
Sciences, 17, 648–665.
Caligiore, D., Borghi, A. M., Parisi, D., & Baldassarre, G. (2010).
TRoPICALS: A computational embodied neuroscience model of
compatibility effects. Psychological Review, 117, 1188–1228.
Caramazza, A., & Shelton, J. (1998). Domain-specific knowledge sys-
tems in the brain: The animate-inanimate distinction. Journal of
Cognitive Neuroscience, 10, 1–34.
Casasanto, D., & Lupyan, G. (2015). All concepts are ad hoc concepts. In
E. Margolis & S. Laurence (Eds.), The conceptual mind: New direc-
tions in the study of concepts (pp. 543–566). Cambridge, MA: MIT
Cognitive Science. Topics in Cognitive Science, 3, 425–437.
Chomsky, N. (1959). A review of B F. Skinner’s Verbal Behavior.
Language, 35, 26–58.
Connell, L., & Lynott, D. (2013). Flexible and fast: Linguistic shortcut
affects both shallow and deep conceptual processing. Psychonomic
Bulletin & Review, 20, 542–550.
Connell, L., & Lynott, D. (2014). Principles of representation: Why you
can’t represent the same concept twice. Topics in Cognitive Science,
Damasio, A. R. (1989). Time-locked multiregional retroactivation: A
systems-level proposal for the neural substrates of recall and recog-
nition. Cognition, 33, 25–62.
Devereux, B. J., Clarke, A., Marouchos, A., & Tyler, L. K. (2013).
Representational similarity analysis reveals commonalities and dif-
ferences in the semantic processing of words and objects. The
Journal of Neuroscience, 33, 18906–18916.
Donald, M. (1993). Precis of Origins of the modern mind:Three stages in
the evolution of culture and cognition.Behavioral and Brain
Sciences, 16, 737–748.
Dove, G. (2009). Beyond perceptual symbols: A call for representational
pluralism. Cognition, 110, 412–431.
Dove, G. (2016). Three symbol ungrounding problems: Abstract con-
cepts and the future of embodied cognition. Psychonomic Bulletin
Drane, D. L., Ojemann, G. A., Aylward, E., Ojemann, J. G., Johnson, L.
C., Silbergeld, D. L., & Tranel, D. (2008). Category-specific naming
and recognition deficits in temporal lobe epilepsy surgical patients.
Neuropsychologia, 46, 1242–1255.
Drane, D. L., Ojemann, G. A., Ojemann, J. G., Aylward, E., Silbergeld,
D. L., Miller, J. W., & Tranel, D. (2009). Category-specific recog-
nition and naming deficits following resection of a right anterior
temporal lobe tumor in a patient with atypical language lateraliza-
tion. Cortex, 45, 630–640.
Drane, D. L., Ojemann, J. G., Phatak, V., Loring, D. W., Gross, R. E.,
Hebb, A. O., Silbergeld, D.L., Miller, J.W., Voets, N.L., Saindane,
A.M., Barsalou, L.W., Meador, K.J., Ojemann, G.A., & Tranel, D.
(2013). Famous face identification in temporal lobe epilepsy:
Support for a multimodal integration model of semantic memory.
Cortex, 49, 1648–1667.
Eliasmith, C. (2013). How to build a brain: A neural architecture for
biological cognition. Oxford: Oxford University Press.
Engel, A. K., Maye, A., Kurthen, M., & König, P. (2013). Where’sthe
action? The pragmatic turn in cognitive science. Trends in Cognitive
Sciences, 17, 202–209.
Erk, K. (2012). Vector space models of word meaning and phrase
meaning: A Survey. Language and Linguistics Compass, 6,
Erk, K., & Padó, S. (2008). A structured vector space model for word
meaning in context. In Proceedings of the Conference on Empirical
Methods in Natural Language Processing (pp. 897–906).
Stroudsburg, PA, USA: Association for Computational Linguistics.
Fairhall, S. L., & Caramazza, A. (2013). Brain regions that represent
amodal conceptual knowledge. Journal of Neuroscience, 33,
Farah, M. J., Wong, A. B., Monheit, M. A., & Morrow, L. A. (1989).
Parietal lobe mechanisms of spatial attention: Modality-specific or
supramodal? Neuropsychologia, 27, 461–470.
Fernandino, L., Binder, J. R., Desai, R. H., Pendl, S. L., Humphries, C. J.,
Gross, W. L., Seidenberg, M. S. (in press). Concept representation
reflects multimodal abstraction: A framework for embodied seman-
tics. Cerebral Cortex.
Fodor, J. A. (1975). The language of thought. Harvard University Press.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive
architecture: A critical analysis. Cognition, 28, 3–71.
Psychon Bull Rev
Ganguli, S., & Sompolinsky, H. (2012). Compressed sensing, sparsity,
and dimensionality in neuronal information processing and data
analysis. Annual Review of Neuroscience, 35, 485–508.
Gawronski, B., & Cesario, J. (2013). Of mice and men What animal
research can tell us about context effects on automatic responses in
humans. Personality and Social Psychology Review, 17, 187–215.
Gibson, J. J. (1979). The ecological approach to visual perception.
Glaser, W. R. (1992). Picture naming. Cognition, 42, 61–105.
Goldinger, S. D., Papesh, M. H., Barnhart, A. S., Hansen, W. A., & Hout,
M. C. (2016). The poverty of embodied cognition. Psychonomic
Bulletin & Review.doi:10.3758/s13423-015-0860-1.
Gotts, S. J. (2016). Incremental learning of perceptual and conceptual
representations and the puzzle of neural repetition suppression.
Psychonomic Bulletin & Review.doi:10.3758/s13423-015-0855-y.
Grill-Spector, K., & Weiner, K. S. (2014). The functional architecture of
the ventral temporal cortex and its role in categorization. Nature
Reviews Neuroscience, 15, 536–548.
Hampton, J. A. (2006). Concepts as prototypes. In B. H. Ross (Ed.), The
psychology of learning and motivation: Advances in research and
theory (Vol 46) (pp. 79–113). San Diego: Elsevier Academic Press.
Harnad, S. (1990). The symbol grounding problem. Physica D:
Nonlinear Phenomena, 42, 335–346.
Hauk, O. (2016). Only time will tell –why temporal information is es-
sential for our neuroscientific understanding of semantics.
Psychonomic Bulletin & Review.doi:10.3758/s13423-015-0873-9.
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people
in the world? Behavioral and Brain Sciences, 33, 61–83.
Hinton, G. E. (2006). Reducing the dimensionality of data with neural
networks. Science, 313, 504–507.
Hsu, N. S., Frankland, S. M., & Thompson-Schill, S. L. (2012).
Chromaticity of color perception and object color knowledge.
Neuropsychologia, 50, 327–333.
Huettig, F., Rommers, J., & Meyer, A. S. (2011). Using the visual world
paradigm to study language processing: A review and critical eval-
uation. Acta Psychologica, 137, 151–171.
Humphries, C., Binder, J., Medler, D., & Liebenthal, E. (2006). Syntactic
and semantic modulation of neural activity during auditory sentence
comprehension. Journal of Cognitive Neuroscience, 18, 665–679.
Jamrozik, A., McQuire, M., Cardillo, E.R., & Chatterjee, A. (2016).
Metaphor: Bridging embodiment to abstraction. Psychonomic
Bulletin & Review.doi:10.3758/s13423-015-0861-0.
Kemmerer, D. (2006). The semantics of space: Integrating linguistic ty-
pology and cognitive neuroscience. Neuropsychologia, 44, 1607–
Kemmerer, D. (2015a). Are the motor features of verb meanings repre-
sented in the precentral motor cortices? Yes, but within the context
of a flexible, multilevel architecture for conceptual knowledge.
Psychonomic Bulletin & Review, 22, 1068–1075.
Kemmerer, D. (2015b). Are we ever aware of concepts? A critical ques-
tion for the Global Neuronal Workspace, Integrated Information,
and Attended Intermediate-Level Representation theories of con-
sciousness. Neuroscience of Consciousness,2015(1), niv006.
Kemmerer, D., Castillo, J. G., Talavage, T., Patterson, S., & Wiley, C.
(2008). Neuroanatomical distribution of five semantic components
of verbs: Evidence from fMRI. Brain and Language, 107, 16–43.
Kiefer, M., Adams, S. C., & Zovko, M. (2012). Attentional sensitization
of unconscious visual processing: Top-down influences on masked
priming. Advances in Cognitive Psychology, 8, 50–61.
Kiefer, M., & Barsalou, L. W. (2013). Grounding the human conceptual
system in perception, action, and internal states. In W. Prinz, M.
Beisert, & A. Herwig (Eds.), Action science: Foundations of an
emerging discipline (pp. 381–407). Cambridge: MIT Press.
Lachman, R., Lachman, J. L., & Butterfield, E. C. (1979). Cognitive
psychology and information processing: An introduction.
Lambon Ralph, M. A., Sage, K., Jones, R. W., & Mayberry, E. J. (2010).
Coherent concepts are computed in the anterior temporal lobes.
Proceedings of the National Academy of Sciences, 107, 2717–2722.
Landauer, T. K., & Dumais, S. T. (1997). A solution to Plato’s problem:
The latent semantic analysis theory of acquisition, induction, and
representation of knowledge. Psychological Review, 104, 211–240.
Landauer, T. K., McNamara, D. S., Dennis, S., & Kintsch, W. (2013).
Handbook of latent semantic analysis. East Sussex: Psychology
Langacker, R. W. (1986). An introduction to cognitive grammar.
Cognitive Science, 10, 1–40.
Langacker, R. W. (2008). Cognitive grammar: A basic introduction.
Oxford: Oxford University Press.
Lebois, L. A. M., Wilson-Mendenhall, C. D., & Barsalou, L. W. (2015).
Are automatic conceptual cores the gold standard of semantic pro-
cessing? The context-dependence of spatial meaning in grounded
congruency effects. Cognitive Science, 39, 1764–1801.
Legrand, D., & Ruby, P. (2009). What is self-specific? Theoretical inves-
tigation and critical review of neuroimaging results. Psychological
Review, 116, 252–282.
Leshinskaya, A., & Caramazza, A. (2016). For a cognitive neuroscience
of concepts: Moving beyond the grounding issue. Psychonomic
Bulletin & Review.doi:10.3758/s13423-015-0870-z.
Louwerse, M. M. (2011). Symbol interdependency in symbolic and em-
bodied cognition. Topics in Cognitive Science, 3, 273–302.
Louwerse, M. M., & Connell, L. (2011). A taste of words: Linguistic
context and perceptual simulation predict the modality of words.
Cognitive Science, 35, 381–398.
Louwerse, M. M., & Jeuniaux, P. (2010). The linguistic and embodied
nature of conceptual processing. Cognition, 114, 96–104.
Love, B. C., Medin, D. L., & Gureckis, T. M. (2004). SUSTAIN: A
network model of category learning. Psychological Review, 111,
Machery, E. (2007). Concept empiricism: A methodological critique.
Cognition, 104, 19–46.
Machery, E. (2016). The amodal brain and the offloading hypothesis.
Psychonomic Bulletin & Review.doi:10.3758/s13423-015-0878-4.
Mahon, B. Z. (2015). What is embodied about cognition? Language,
Cognition and Neuroscience, 30, 420–429.
Mahon, B. Z., & Caramazza, A. (2008). A critical look at the embodied
cognition hypothesis and a new proposal for grounding conceptual
content. Journal of Physiology-Paris, 102, 59–70.
Margolis, E., & Laurence, S. (1999). Concepts: Core readings. MIT
Martin, A. (2007). The representation of object concepts in the brain.
Annual Review of Psychology, 58, 25–45.
Martin, A. (2009). Circuits in mind: The neural foundations for object
concepts. In M. Gazzaniga (Ed.), The cognitive neurosciences
(4th ed., pp. 1031–1045). Cambridge, MA: MIT Press.
Martin, A. (2016). GRAPES—Grounding representations in action, per-
ception, and emotion systems: How object properties and categories
are represented in the human brain. Psychonomic Bulletin & Review.
Martin, A., & Chao, L. L. (2001). Semantic memory and the brain: struc-
ture and processes. Current Opinion in Neurobiology, 11, 194–201.
Martin, A., Simmons, W. K., Beauchamp, M. S., & Gotts, S. J. (2014). Is
asingle‘hub’, with lots of spokes, an accurate description of the
neural architecture of action semantics? Physics of Life Reviews, 11,
McClelland, J. L., & Rumelhart, D. E. (1985). Distributed memory and
the representation of general and specific information. Journal of
Experimental Psychology: General, 114, 159–188.
McRae, K., Cree, G. S., Seidenberg, M. S., & McNorgan, C. (2005).
Semantic feature production norms for a large set of living and
nonliving things. Behavior Research Methods, 37, 547–559.
Psychon Bull Rev
McRae, K., & Jones, M. N. (2013). Semantic memory. In D. Reisberg
(Ed.), The Oxford Handbook of Cognitive Psychology (pp. 206–
219). Oxford: Oxford University Press.
Meyer, K., & Damasio, A. (2009). Convergence and divergence in a
neural architecture for recognition and memory. Trends in
Neurosciences, 32, 376–382.
Mitchell, T. M., Shinkareva, S. V., Carlson, A., Chang, K.-M., Malave, V. L.,
Mason, R. A., & Just, M. A. (2008). Predicting human brain activity
associated with the meanings of nouns. Science, 320, 1191–1195.
Moors, A., & De Houwer, J. (2006). Automaticity: A theoretical and
conceptual analysis. Psychological Bulletin, 132, 297–326.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: MIT
Murphy, G. L. (2016). Is there an exemplar theory of concepts?
Psychonomic Bulletin & Review.doi:10.3758/s13423-015-0834-3.
Niedenthal, P. M., Mermillod, M., Maringer, M., & Hess, U. (2010). The
Simulation of Smiles (SIMS) Model: Embodied simulation and the
meaning of facial expression. Behavioral and Brain Sciences, 33,
O’Regan, J. K., & Noë, A. (2001). A sensorimotor account of vision and
visual consciousness. Behavioral and Brain Sciences, 24, 939–973.
Padó, S., & Lapata, M. (2007). Dependency-based construction of se-
mantic space models. Computational Linguistics, 33, 161–199.
Paivio, A. (1986). Mental representations: A dual-coding approach.
Oxford: Oxford University Press.
Papies, E. K. (2013). Tempting food words activate eating simulations.
Frontiers in Psychology, 4, 1–12.
Papies, E. K., & Barsalou, L. W. (2015). Grounding desire and motivated
behavior: A theoretical framework and review of empirical evi-
dence. In W. Hofmann & L. F. Nordgren (Eds.), The psychology
of desire (pp. 36-60). New York: Guilford Press.
Patterson, K., Nestor, P. J., & Rogers, T. T. (2007). Where do you know
what you know? The representation of semantic knowledge in the
human brain. Nature Reviews Neuroscience, 8, 976–987.
Pulvermüller, F. (2012). Meaning and the brain: The neurosomatics of
referential, interactive, and combinatorial knowledge. Journal of
Neurolinguistics, 25, 423–459.
Pulvermüller, F. (2013). How neurons make meaning: brain mechanisms
for embodied and abstract-symbolic semantics. Trends in Cognitive
Sciences, 17, 458–470.
Pulvermüller, F., & Fadiga, L. (2010). Active perception: sensorimotor
circuits as a cortical basis for language. Nature Reviews
Neuroscience, 11, 351–360.
Pulvermüller, F., & Garagnani, M. (2014). From sensorimotor learning to
memory cells in prefrontal and temporal association cortex: A
neurocomputational study of disembodiment. Cortex, 57, 1–21.
Pylyshyn, Z. W. (1984). Computation and cognition. Cambridge: MIT Press.
Reilly, J., Peelle, J. A., Garcia, A., & Crutch, S. J. (2016). Linking somatic
and symbolic representation in semantic memory: the dynamic mul-
tilevel reactivation framework. Psychonomic Bulletin & Review.doi:
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel
distributed processing approach. Cambridge: MIT Press.
Rosch, E., & Mervis, C. B. (1975). Family resemblances: Studies in the
internal structure of categories. Cognitive Psychology, 7, 573–605.
Ross, B. H., & Murphy, G. L. (1999). Food for thought: Cross-
classification and category organization in a complex real-world
domain. Cognitive Psychology, 38, 495–553.
Rudebeck, P. H., & Murray, E. A. (2014). The orbitofrontal oracle:
Cortical mechanisms for the prediction and evaluation of specific
behavioral outcomes. Neuron, 84, 1143–1156 .
Rumiati, R. I., & Foroni, F. (2016). We are what we eat: How food is
represented in our mind/brain. Psychonomic Bulletin & Review.doi:
Santiago, J., Román, A., & Ouellet, M. (2011). Flexible foundations of
abstract thought: A review and a theory. In A. Maass & T. W.
Schubert (Eds.), Spatial dimensions of social thought (pp. 41–
110). Mouton de Gruyter: Berlin.
Schrodt, F., Layher, G., Neumann, H., & Butz, M. V. (2015). Embodied
learning of a generative neural model for biological motion percep-
tion and inference. Frontiers in Computational Neuroscience, 9, 79.
Schwanenflugel, P. J. (1991). Why are abstract concepts so hard to un-
derstand? In P. J. Schwanenflugel (Ed.), The psychology of word
meanings (pp. 223–250). Hillsdale: Lawrence Erlbaum Associates.
Schwiedrzik, C. M., Bernstein, B., & Melloni, L. (2016). Motion along
the mental number line reveals shared representations for
numerosity and space. eLife,5.
Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain
Sciences, 3, 417–424.
Simmons, W. K., & Barsalou, L. W. (2003). The similarity-in-topography
principle: Reconciling theories of conceptual deficits. Cognitive
Neuropsychology, 20, 451–486.
Simmons, W. K., Ramjee, V., Beauchamp, M. S., McRae, K., Martin, A.,
& Barsalou, L. W. (2007). A common neural substrate for perceiv-
ing and knowing about color. Neuropsychologia, 45, 2802–2810.
Smith, E. E., & Medin, D. L. (1981). Categories and concepts.
Cambridge: Harvard University Press.
Sporns, O. (Ed.). (2010). Analysis and function of large-scale brain
networks. Washington, DC: Society for Neuroscience.
Talmy, L. (1985). Lexicalization patterns: Semantic structure in lexical
forms. Language Typology and Syntactic Description, 3, 57–149.
Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy,
J. C. (1995). Integration of visual and linguistic information in spo-
ken language comprehension. Science, 268, 1632–1634.
Thagard, P., & Stewart, T. C. (2011). The AHA! experience: Creativity
through emergent binding in neural networks. Cognitive Science,
Tomasino, B., & Rumiati, R. I. (2013). At the mercy of strategies: The
role of motor representations in language understanding. Frontiers
in Psychology, 4.
van Dam, W. O., Brazil, I. A., Bekkering, H., & Rueschemeyer, S.-A.
(2014). Flexibility in embodied language processing: Context ef-
fects in lexical access. Topics in Cognitive Science, 6, 407–424.
Van Doren, L., Dupont, P., De Grauwe, S., Peeters, R., & Vandenberghe, R.
(2010). The amodal system for conscious word and picture identifi-
cation in the absence of a semantic task. NeuroImage, 49, 3295–3307.
Van Opstal, F., & Verguts, T. (2013). Is there a generalized magnitude
system in the brain? Behavioral, neuroimaging, and computational
evidence. Frontiers in Psychology,4.
Waldhauser, G. T., Braun, V., & Hanslmayr, S. (2016). Episodic
memory retrieval functionally relies on very rapid reactiva-
tion of sensory information. The Journal of Neuroscience, 36,
Walsh, V. (2003). A theory of magnitude: common cortical metrics of
time, space and quantity. Trends in Cognitive Sciences, 7, 483–488.
Wang, X., Han, Z., He, Y., Caramazza, A., Song, L., & Bi, Y. (2013).
Where color rests: Spontaneous brain activity of bilateral fusiform
and lingual regions predicts object color knowledge performance.
NeuroImage, 76, 252–263.
Wilson-Mendenhall, C. D., Simmons, W. K., Martin, A., & Barsalou, L.
W. (2013). Contextual processing of abstract concepts reveals neural
representations of nonlinguistic semantic content. Journal of
Cognitive Neuroscience, 25, 920–935.
Wilson, R. C., Takahashi, Y. K., Schoenbaum, G., & Niv, Y. (2014).
Orbitofrontal cortex as a cognitive map of task space. Neuron, 81,
Wong, C., & Gallate, J. (2012). The function of the anterior temporal
lobe: A review of the empirical evidence. Brain Research, 1449,
Yates, M. J., Loetscher, T., & Nicholls, M. E. R. (2012). A generalized
magnitude system for space, time, and quantity? A cautionary note.
Psychon Bull Rev
Yee, E., & Thompson-Schill, S. L. (2016). Putting concepts into context.
Psychonomic Bulletin & Review.doi:10.3758/s13423-015-0948-7.
Zwaan, R. A. (2004). The immersed experiencer: Toward an embodied
theory of language comprehension. In B. H. Ross (Ed.), The
Psychology of Learning and Motivation: Advances in Research
and Theory (Vol. 44, pp. pp. 35–pp. 62). San Diego: Elsevier
Zwaan, R. A. (2016). Situation models, mental simulations, and abstract
concepts in discourse comprehension. Psychonomic Bulletin &
Zwaan, R. A., & Madden, C. J. (2005). Embodied sentence comprehen-
sion. In D. Pecher & R. A. Zwaan (Eds.), Grounding cognition: The
role of perception and action in memory, language, and thinking
(pp. 224–245). New York: Cambridge University Press.
Psychon Bull Rev