-
Mirror neurons: From origin tofunction
Richard CookDepartment of Psychology, City University London,
London EC1R 0JD,United Kingdom
[email protected]://www.city.ac.uk/people/academics/richard-cook
Geoffrey BirdMRC Social, Genetic and Developmental Psychiatry
Centre, Institute of
Psychiatry, Kings College London, London SE5 8AF, United
[email protected]://sites.google.com/site/geoffbirdlab/http://www.iop.kcl.ac.uk/staff/prole/default.aspx?go=13152
Caroline CatmurDepartment of Psychology, University of Surrey,
Guildford, Surrey GU2 7XH,
United
[email protected]://www.surrey.ac.uk/psychology/people/dr_caroline_catmur/http://sites.google.com/site/carolinecatmur/
Clare PressDepartment of Psychological Sciences, Birkbeck
College, University ofLondon, London WC1E 7HX, United Kingdom
[email protected]://www.bbk.ac.uk/psychology/actionlab/http://www.bbk.ac.uk/psychology/our-staff/academic/dr-clare-press
Cecilia HeyesAll Souls College, University of Oxford, Oxford,
OX1 4AL, and Department ofExperimental Psychology, University of
Oxford, Oxford OX1 3UD, UnitedKingdom
[email protected]://www.all-souls.ox.ac.uk/users/heyesc/
Abstract: This article argues that mirror neurons originate in
sensorimotor associative learning and therefore a new approach is
needed toinvestigate their functions. Mirror neurons were
discovered about 20 years ago in the monkey brain, and there is now
evidence that theyare also present in the human brain. The
intriguing feature of many mirror neurons is that they re not only
when the animal isperforming an action, such as grasping an object
using a power grip, but also when the animal passively observes a
similar actionperformed by another agent. It is widely believed
that mirror neurons are a genetic adaptation for action
understanding; that theywere designed by evolution to fulll a
specic socio-cognitive function. In contrast, we argue that mirror
neurons are forged bydomain-general processes of associative
learning in the course of individual development, and, although
they may have psychologicalfunctions, they do not necessarily have
a specic evolutionary purpose or adaptive function. The evidence
supporting this view showsthat (1) mirror neurons do not
consistently encode action goals; (2) the contingency- and
context-sensitive nature of associativelearning explains the full
range of mirror neuron properties; (3) human infants receive enough
sensorimotor experience to supportassociative learning of mirror
neurons (wealth of the stimulus); and (4) mirror neurons can be
changed in radical ways bysensorimotor training. The associative
account implies that reliable information about the function of
mirror neurons can be obtainedonly by research based on
developmental history, system-level theory, and careful
experimentation.
Keywords: action understanding; associative learning; contextual
modulation; contingency; genetic adaptation; imitation; mirror
neuron;poverty of the stimulus; sensorimotor experience.
1. Introduction
Mirror neurons (MNs) were discovered serendipitously in1992 and
given their brilliant name four years later (di Pel-legrino et al.
1992; Gallese et al. 1996). The striking feature
of many MNs is that they re not only when a monkey isperforming
an action, such as grasping an object using apower grip, but also
when the monkey passively observesa similar action performed by
another. Neurons with thiscapacity to match observed and executed
actions, to code
BEHAVIORAL AND BRAIN SCIENCES (2014) 37,
177241doi:10.1017/S0140525X13000903
Cambridge University Press 2014 0140-525X/14 $40.00 177
-
both my action and your action, were originally found inarea F5
of the ventral premotor cortex (PMC) (di Pelle-grino et al. 1992;
Gallese et al. 1996) and the inferior par-ietal lobule (IPL)
(Bonini et al. 2010; Fogassi et al. 2005) of
the monkey brain. There is now a substantial body of evi-dence
suggesting that MNs are also present in the humanbrain (Molenberghs
et al. 2012).MNs have received a great deal of attention from
special-
ists and in the scientic and public media. Hailed as cellsthat
read minds (Blakesee 2006), the neurons that shapedcivilization
(Ramachandran 2009), and a revolution inunderstanding social
behavior (Iacoboni 2008), MNs havebeen ascribed a wide variety of
functions. The primary can-didates relate to action understanding
(Gallese & Sinigaglia2011; Rizzolatti et al. 1996), imitation
(Iacoboni et al.1999), and language processing (Rizzolatti &
Arbib 1998).However, signifying the way in which MNs have
capturedthe attention and imagination of neuroscientists,
psycholo-gists, and philosophers, they have also been implicated
in:embodied simulation (Aziz-Zadeh et al. 2006b), empathy(Avenanti
et al. 2005), emotion recognition (Enticottet al. 2008),
intention-reading (Iacoboni et al. 2005),language acquisition
(Theoret & Pascual-Leone 2002),language evolution (Arbib 2005),
manual communication(Rizzolatti et al. 1996), sign language
processing (Corina& Knapp 2006), speech perception (Glenberg et
al.2008), speech production (Kuhn & Brass 2008), music
pro-cessing (Gridley & Hoff 2006), sexual orientation
(Ponsetiet al. 2006), and aesthetic experience (Cinzia &
Gallese2009). In addition, it has been suggested that MN
dysfunc-tion contributes to a number of disorders, including
autism(Dapretto et al. 2006; Nishitani et al. 2004; J. H.
Williamset al. 2001), schizophrenia (Arbib & Mundhenk
2005),Downs syndrome (Virji-Babul et al. 2008), multiple scler-osis
(Rocca et al. 2008), cigarette addiction (Pineda &Oberman
2006), and obesity (Cohen 2008).Thus, much of the rst 20 years of
MN research has been
devoted to theorizing and speculation about their func-tions. In
contrast, the primary focus of this article is theorigin of MNs.
Our principal questions are not What doMNs do? or What are they
for?, but What is theprocess that gives MNs their mirrorness; their
fascinating,cardinal capacity to match observed with
executedactions?The standard view of MNs, which we will call
the
genetic account, alloys a claim about the origin of MNswith a
claim about their function. It suggests that the mir-rorness of MNs
is due primarily to heritable genetic factors,and that the genetic
predisposition to developMNs evolvedbecause MNs facilitate action
understanding. In the senseof an adaptation developed by G. C.
Williams, and usedin Evolutionary Psychology, the genetic account
castsMNs as an adaptation for action understanding. In contrast,we
argue in this article that the balance of evidence cur-rently
favors an associative account of MNs, which separ-ates questions
about their origin and function. It suggeststhat MNs acquire their
capacity to match observed withexecuted actions through
domain-general processes of sen-sorimotor associative learning, and
that the role of MNs inaction understanding, or any other social
cognitive func-tion, is an open empirical question. The
associativeaccount is functionally permissive; it allows, but does
notassume, that MNs make a positive contribution to
socialcognition. Thus, there are three critical differencesbetween
the genetic and associative accounts: (1) Theformer combines, and
the latter dissociates, questionsabout origin and function. (2) The
genetic account sug-gests that natural selection has acted directly
on MNs,
RICHARD COOK completed his Ph.D. in the Cognitive,Perceptual and
Brain Sciences Department at Univer-sity College London in 2012,
and took up a lectureshipat City University London in the same
year. Hisresearch, which combines the methods of social cogni-tive
neuroscience and visual psychophysics, examinesthe mechanisms of
imitation and the perceptual pro-cesses recruited by faces, bodies,
and actions. Hiswork has been published in the Journal of
ExperimentalPsychology, the Journal of Vision, Proceedings of
theRoyal Society, and Psychological Science.
GEOFF BIRD is a Senior Lecturer at the MRC (MedicalResearch
Council) Social, Genetic, and DevelopmentalPsychiatry Centre at the
Institute of Psychiatry, KingsCollege London. After being taught
how to thinkby Cecilia Heyes during his Ph.D. studies at
UniversityCollege London, Geoff now studies typical and
atypicalsocial cognition in collaboration with (among others)
hisexcellent co-authors on this target article.
CAROLINE CATMUR is an ESRC (Economic and SocialResearch Council)
Future Research Leader and SeniorLecturer in Cognitive Psychology
at the University ofSurrey. She received her B.A. in 2002 from the
Univer-sity of Oxford and her Ph.D. in 2009 from UniversityCollege
London. Her research investigates the cognitiveand neural
mechanisms underlying social cognition. Hercurrent work, which
combines behavioural studies withneuroimaging and transcranial
magnetic stimulation,focuses on the development and control of
social cogni-tive processes, including imitation,
perspective-taking,and empathy.
CLARE PRESS is a Lecturer in Psychological Sciences atBirkbeck,
University of London, UK. She received herB.Sc. and Ph.D. from
University College London, andsubsequently worked as an ESRC/MRC
PostdoctoralResearch Fellow at the Wellcome Trust Centre
forNeuroimaging. Her research uses behavioural and neu-roimaging
techniques, with a focus on electrophysiologi-cal methods, to
examine the mechanisms underlyingperception-action mapping for
social processes such asimitation and action perception, as well as
action selec-tion and control functions.
CECILIA HEYES is a Senior Research Fellow in Theor-etical Life
Sciences and full Professor of Psychology atAll Souls College,
University of Oxford, and a Fellowof the British Academy. Her work
concerns the evol-ution of cognition. It explores the ways in
whichnatural selection, learning, and developmental and cul-tural
processes combine to produce the mature cogni-tive abilities found
in adult humans. Her experimentaland theoretical research,
initially in animal cognitionand later in human cognitive
neuroscience, hasfocused on imitation, social learning,
self-recognition,and theory of mind. She is currently asking to
whatextent these processes of cultural learning are, like
lit-eracy, culturally inherited.
Cook et al.: Mirror neurons: From origin to function
178 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
whereas the associative account suggests that natural selec-tion
has played a background role; for example, acting ondomain-general
mechanisms of associative learning. (3)The genetic account assigns
a relatively minor, facilitativerole to sensory and/or motor
experience in the develop-ment of MNs, whereas the associative
account suggeststhat sensorimotor experience plays a major,
instructiverole in their development.We begin, in section 2, with
some basic information
about the ways in which MNs have been dened and inves-tigated in
both monkeys and humans.1 In the third sectionwe present the
genetic and associative accounts, and intro-duce four kinds of
evidence that have the potential to favorone of these hypotheses
over the other. Sections 47discuss each of these types of evidence
in turn. In section8 we survey recent theories that are, or appear
to be,alternatives to the genetic and associative accounts,
andsuggest that the associative account is stronger. Finally,
insection 9 we argue that the associative account has
majormethodological implications for research investigating
thefunctions of MNs. Unlike the genetic account, the associat-ive
account doesnt claim to tell us what MNs do or whatthey are for,
but it does tell us how we can nd out.
2. Mirror neuron basics
2.1. Locations and denitions
MNs have been found in the monkey brain (Macaca nemis-trina and
Macaca mulatta), not only in classical areas ventral PMC and IPL
but also in non-classical areas,including primary motor cortex
(Dushanova & Donoghue2010; Tkach et al. 2007) and dorsal PMC
(Tkach et al.2007). There is also evidence of single neurons,
orcircumscribed populations of neurons, with sensorimotormatching
properties in classical areas of the humanbrain, including
posterior regions of the inferior frontalgyrus (IFG; considered the
human homologue of themonkey F5) (Kilner et al. 2009) and inferior
parietalcortex (Chong et al. 2008), and non-classical areas of
thehuman brain, including dorsal PMC, superior parietallobule, and
cerebellum (Molenberghs et al. 2012), supple-mentary motor area,
and medial temporal lobe (Mukamelet al. 2010).Some researchers
apply the term mirror neuron only to
neurons found in classical areas (Brown & Brune
2012;Molenberghs et al. 2012), whereas others, like us, use theterm
to refer to neurons in both classical and non-classicalareas
(Gallese & Sinigaglia 2011; Keysers & Gazzola 2010).In
addition to this variation in anatomical specicity, someresearchers
reserve the term mirror neuron for units thatdischarge during the
observation and execution of precisely(Dinstein et al. 2008b;
Keysers 2009) or broadly similaractions (Kilner et al. 2009),
whereas others use the term,at least on occasions, to refer to any
neuron that is respon-sive to both the observation and execution of
action,regardless of whether the observed and executed actionsare
even broadly similar to one another (Gallese et al.1996; Rizzolatti
& Craighero 2004). In accord with themajority of researchers in
the eld, and the meaning ofthe word mirror, we take it to be a
cardinal feature ofMNs that they are responsive to observation and
executionof similar actions. However, following common usage,
wealso refer to logically related MNs (see sect. 2.2), which
re during observation and execution of dissimilaractions, as
mirror neurons.
2.2. Monkeys
Early studies of the eld properties of monkey MNs thesensory and
motoric conditions in which they re revealed three basic types:
Strictly congruent MNs dis-charge during observation and execution
of the sameaction, for example, precision grip. Broadly
congruentMNs are typically active during the execution of oneaction
(e.g., precision grip) and during the observation ofone or more
similar, but not identical, actions (e.g.,power grip alone, or
precision grip, power grip, and grasp-ing with the mouth).
Logically related MNs respond todifferent actions in observe and
execute conditions. Forexample, they re during the observation of
an exper-imenter placing food in front of the monkey, and whenthe
monkey grasps the food in order to eat it. MNs donot respond to the
presentation of objects alone (di Pelle-grino et al. 1992).
However, canonical neurons, whichare active during object
observation and performance ofan action that is commonly performed
on that object,are co-located with MNs both in area F5 (Murata et
al.1997) and in the anterior intraparietal sulcus (Murataet al.
2000).To date, monkey MNs have been found that are respon-
sive to the observation and execution of hand and mouthactions.
The hand actions include grasping, placing, manip-ulating with the
ngers, and holding (di Pellegrino et al.1992; Gallese et al. 1996).
The mouth actions include inges-tive behaviors such as breaking
food items, chewing andsucking, and communicative gestures such as
lip-smacking,lip-protrusion, and tongue-protrusion (Ferrari et al.
2003).
2.3. Humans
Only one study purports to offer direct evidence fromsingle cell
recording of MNs in the human brain(Mukamel et al. 2010). However,
there is a considerablebody of indirect evidence from neuroimaging,
transcra-nial magnetic stimulation (TMS), and behavioral studies
suggesting that human brains contain MNs or comparablemirror
mechanisms; circumscribed cortical areasinvolved in both action
production and observation (Glen-berg 2011).
2.3.1. Neuroimaging. Functional magnetic resonanceimaging (fMRI)
has identied regions of PMC (bothclassic BA6 and BA44) and inferior
parietal areas that areactive during both action observation and
execution(Aziz-Zadeh et al. 2006a; Buccino et al. 2004; Carr et
al.2003; Grzes et al. 2003; Iacoboni et al. 1999; Leslieet al.
2004; Tanaka & Inui 2002; Vogt et al. 2007). Overlap-ping
responses to action observation and execution havebeen found in
single-subject analyses of unsmoothed data(Gazzola & Keysers
2009), conrming that the foregoingreports are not artifacts of
group averaging. Most recently,repetition suppression protocols
have been used to provideevidence of mirror populations encoding
visual and motorrepresentations of the same action. These
paradigmsexploit the logic that repeated stimulus presentation
oraction execution causes a decrease in neural
responses(Grill-Spector et al. 2006). Cross-modal repetition
Cook et al.: Mirror neurons: From origin to function
BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2 179
-
suppression effects have been reported, whereby
actionobservation followed by execution of the same action, orvice
versa, elicits repetition suppression in inferior parietalregions
(Chong et al. 2008; Lingnau et al. 2009) and inPMC (Kilner et al.
2009; Lingnau et al. 2009).
2.3.2. Mirror pattern of MEPs. A human mirror mechanismis
further suggested by mirror motor evoked potentials(MEPs) elicited
during action observation (Fadiga et al.1995). When TMS is applied
to M1 during passiveaction observation, the amplitude of the MEPs
recordedfrom the muscles required to execute that action isgreater
than the amplitude of the MEPs recorded whenobserving a different
action. For example, observingindex and little nger abduction
movements selectivelyfacilitates the amplitude of MEPs recorded
from the rstdorsal interosseus and abductor digiti minimi
muscles,responsible for index and little nger movements,
res-pectively (Catmur et al. 2011). That action
observationselectively increases corticospinal excitability to
actionrelevant muscles is suggestive of mirror
sensorimotorconnectivity.
2.3.3. Automatic imitation. Automatic imitation is said tooccur
when observation of an action involuntarily facilitatesperformance
of a topographically similar action (body partsmake the same
movements relative, not to external framesof reference, but to one
another) and/or interferes withperformance of a topographically
dissimilar action (Brasset al. 2001; Strmer et al. 2000). Humans
show robust auto-matic imitation when they observe hand, arm, foot,
andmouth movements (Heyes 2011). This is regarded bymany
researchers as evidence of a human mirror mechan-ism (Blakemore
& Frith 2005; Ferrari et al. 2009a; Iaco-boni 2009; Kilner et
al. 2003; Longo et al. 2008; vanSchie et al. 2008). Supporting this
view, several studieshave shown that application of disruptive TMS
to theIFG a classical mirror area interferes with
automaticimitation (Catmur et al. 2009; Newman-Norlund et
al.2010).
3. The mirrorness of mirror neurons: Genetic orassociative?
This section presents the standard, genetic account of theorigin
of MNs and the alternative associative account.
3.1. Genes for mirroring?
The genetic account assumes: (1) Among common ances-tors of
extant monkeys and humans, some individuals hada stronger genetic
predisposition to develop MNs, and(2) these individuals were more
reproductively successfulthan those with a weaker genetic
predisposition becausethe development of MNs enhanced their
capacity to under-stand the actions of other agents. Consequently,
(3) agenetic predisposition to develop MNs became universal,or
nearly universal, in monkeys and humans. (4) Motorexperience (the
performance of actions) and/or sensoryexperience (the observation
of actions) plays a facilitative(Gottlieb 1976) or permissive
(Gilbert 2003) role in thedevelopment of MNs, but their matching
properties areprimarily due to this genetic predisposition.
The term action understanding was introduced by Riz-zolatti and
colleagues to characterize the function of MNs(Rizzolatti &
Fadiga 1998; Rizzolatti et al. 1996). As far aswe are aware, it had
not previously been used in researchon animal or human cognition.
The term plays a key rolein the genetic account; it describes the
adaptive functionof MNs, the effects that made them a target of
positiveselection pressure. However, there is still no
consensusabout exactly what is meant by action understanding, orhow
it differs from cognate functions such as action per-ception,
action recognition, and action selection(Gallese et al. 2011).
Attempts to clarify have emphasizedthat, in comparison with purely
visual processing ofaction, MN activity relates to the meaning of
an actionand yields a richer understanding, real understanding,or
understanding from within (Gallese et al. 2011; Rizzo-latti &
Sinigaglia 2010). As we discuss further in section 8,these
descriptions do not provide an operational denitionof action
understanding, that is, a denition that wouldallow behavior based
on (this kind of) action understandingto be distinguished
empirically from behavior based onother processes.Until recently,
the genetic account was largely implicit in
discussions of the evolution of MNs (Gallese & Goldman1998;
Rizzolatti & Arbib 1998; Rizzolatti & Craighero 2004;M. J.
Rochat et al. 2010). For example, it has beensuggested that the
mirror neuron mechanism is a mechan-ism of great evolutionary
importance through which pri-mates understand actions done by their
conspecics(Rizzolatti & Craighero 2004, p. 172) and that in
theirbasic properties, MNs constitute a relatively
simpleaction-perception mechanism that could have beenexploited
several times in the course of animal evolution(Bonini &
Ferrari 2011, p. 172). A number of discussionshave also expressed
the view that MNs are present atbirth (Ferrari et al. 2009; Gallese
et al. 2009; Lepage &Theoret 2007; Rizzolatti & Fadiga
1998), a feature com-monly associated with traits for which there
is stronggenetic predisposition (Mameli & Bateson 2006).
Forexample, Casile and colleagues have suggested that bothface
processing and the mirror neuron system, or at leastthe part
involved in facial movements, rely on a brainnetwork that is
present already at birth and whose elementsare probably genetically
predetermined (Casile et al. 2011,p. 531).In its starkest form, the
genetic hypothesis would suggest
that gene-based natural selection has provided each indi-vidual
monkey and human with MNs that code themapping between a xed set of
observed and executedactions, and that experience plays a minimal
role in thedevelopment of the observation-execution matching
prop-erties of these neurons. However, the genetic hypothesisdoes
not necessarily assume that experience plays aminimal role. For
example, in a recent explicit statementof the genetic account,
Gallese et al. (2009) suggestedthat links form during gestation
between motor regionsand to-become-visual regions that will
subsequentlymediate sensorimotor matching abilities in young
infants.They implied that these projections are genetically
predis-posed to target certain visual areas, and therefore that
thematching properties of MNs are produced by informationencoded in
the genome. However, they also suggestedthat motor experience plays
a part in preparing motorregions to send projections to visual
areas, and that visual
Cook et al.: Mirror neurons: From origin to function
180 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
experience may also facilitate the maturation of fully
func-tioning MNs.
3.2. A product of associative learning
The associative hypothesis assumes that gene-based
naturalselection has played a signicant background role withrespect
to the development of MNs; for example, inshaping the anatomy of
visual and motor cortex for visualguidance of action, and in
producing the capacity for associ-ative learning in neural tissue.
However, it suggests that thecardinal matching properties of MNs
are a product, not of aspecic genetic predisposition, but of
domain-general pro-cesses of associative learning the same kind of
learningthat produces Pavlovian and instrumental
conditioningphenomena (Catmur et al. 2009; Heyes 2010; Ray
&Heyes 2011). Associative learning is found in a widerange of
vertebrate and invertebrate species, indicatingthat it is an
evolutionarily ancient and highly conservedadaptation for tracking
predictive relationships betweenevents (Heyes 2012b; Schultz &
Dickinson 2000).Figure 1 is a schematic representation of howMNs
could
acquire their matching properties through
sensorimotorassociative learning. Before associative learning,
sensoryneurons in the superior temporal sulcus (STS), responsiveto
different high-level visual properties of observed action(Oram
& Perrett 1994; 1996), are weakly connected,directly or
indirectly, to motor neurons in PMC (Rizzolattiet al. 1988) and
parietal cortex (Gallese et al. 2002). Someof these connections may
be stronger than others, but thelinks between sensory and motor
neurons coding similar
actions are not consistently stronger than other, non-matching
links. The kind of learning that produces MNsoccurs when there is
correlated (i.e., contiguous and con-tingent) excitation of sensory
neurons and motor neuronsthat code similar actions. For example,
when an adult imi-tates an infants facial movements, there might be
corre-lated excitation of neurons that are responsive to
theobservation and execution of lip protrusion.
Correlatedexcitation of the sensory and motor neurons increases
thestrength of the connection between them, so that sub-sequent
excitation of the sensory neuron propagates tothe motor neuron.
Thereafter, the motor neuron res,not only during execution of lip
protrusion, but also, viaits connection with the sensory neuron,
during observationof lip protrusion; what was originally a motor
neuron hasbecome a lip protrusion MN. Correlated excitation
ofsensory and motor neurons encoding the same propertyof action
occurs not only when humans are imitated, butalso when we observe
our own actions (directly or usingan optical mirror); observe
others during the kind of syn-chronous activities involved in
sports and dance training;and as a consequence of acquired
equivalence experi-ence, for example, when the same sound (a word,
or asound produced by an action) is paired sometimes
withobservation of an action and sometimes with its
execution(Catmur et al. 2009; Ray & Heyes 2011).There are
several important things to note about the
associative hypothesis:
1. Strong experience-dependence It suggests thatcorrelated
sensorimotor experience plays an inductive
Figure 1. Mirror neurons from associative learning. (a) Before
learning, sensory neurons in STS, encoding visual descriptions
ofobserved action, are not systematically connected to motor
neurons in premotor and parietal areas involved in the production
ofsimilar actions. (b) Through social interaction and
self-observation in the course of typical development, agents
receive correlatedsensorimotor experience; they see and do the same
action at about the same time (contiguity), with one event
predicting the other(contingency). This experience produces
correlated activation of sensory and motor neurons coding similar
actions, and, throughassociative learning, (c) strengthens
connections between these neurons. Due to these connections,
neurons that were once involvedonly in the execution of action will
also discharge during observation of a similar action; motor
neurons become MNs (see sect. 3.2).Because the visual system and
motor system are organised hierarchically, some types of
sensorimotor experience produce correlatedactivation of sensory and
motor neurons coding relatively low-level features of action (e.g.,
left or right hand, power or precisiongrip), and thereby generate
strictly congruent, hand- and direction-sensitive MNs. Other types
produce correlated activation ofneurons coding relatively
high-level features (e.g., grasping) and generate broadly congruent
MNs (see sect. 5.1).
Cook et al.: Mirror neurons: From origin to function
BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2 181
-
(Gottlieb 1976) or instructive (Gilbert 2003) role; withoutthis
kind of experience, MNs would not develop at all.2. Social
construction It proposes that much of the
sensorimotor experience required for MN developmentcomes from
being imitated, synchronous action, andexposure to action words
(Ray & Heyes 2011), and there-fore that MNs are to a very large
extent built throughsocial interaction.3. Contingency Following
contemporary learning
theory, the associative account species that MN develop-ment
requires, not just that sensory and motor neuronsre together in a
Hebbian way (contiguity), but that theevent provoking ring of one
predicts the event provokingring of the other (contingency; Cook et
al. 2010).4. Testability The associative account makes novel
predictions about the development and mature propertiesof MNs,
many of which have already been tested andsupported by experiments
using a variety of methods (seesect. 7).Thus, the associative
hypothesis implies that the cha-
racteristic, matching properties of MNs result from agenetically
evolved process, associative learning, but thisprocess was not
designed by genetic evolution toproduce MNs. Rather, it just
happens to produce MNswhen the developing system receives
correlated experienceof observing and executing similar actions.
When thesystem receives correlated experience of observingobjects
and executing actions, the same associativeprocess produces
canonical neurons. When the systemreceives correlated experience of
observing one actionand executing a different action, the same
associativeprocess produces logically related MNs.
3.3. Not nature versus nurture
The contrast between the genetic and associative hypoth-eses
does not represent a dichotomous naturenurturedebate. It has been
recognized for decades that the devel-opment of all phenotypic
characteristics depends on theinteraction of nature and nurture,
genes and the environ-ment, evolution and learning (Elman et al.
1996; Oyama1985). Rather, the two accounts differ in the
specicroles they assign to genetic evolution and to learning, andin
the types of experience they take to be important, in pro-ducing
the characteristic matching properties of MNs. Thegenetic
hypothesis says that genetic evolution has played aspecic and
decisive role, and learning based on sensoryand/or motor experience
plays a merely facilitative role,in the development of MNs. In
contrast, the associativehypothesis says that genetic evolution has
played a non-specic background role, and that the characteristic
match-ing properties of MNs are forged by sensorimotor
learning.Regarding the function of MNs, the genetic account
assumes that they play a fundamental role in action
under-standing, and that this is why a specic genetic
predisposi-tion to develop MNs was favored by natural selection.
Inother words, it proposes that action understanding is theadaptive
function of MNs, or that MNs are an adap-tation for action
understanding. In this way, the geneticaccount offers a hypothesis
about the function of MNs asan explanation for their origins. In
contrast, the associativeaccount separates questions about the
origin and functionof MNs. It suggests that MNs develop through
associativelearning, and that further research is needed to nd
out
how they contribute to social cognition (see sect. 9). Ifthis
research reveals that MNs make positive contributionsto social
cognition, these would be psychological uses orpsychological
functions, but not necessarily adaptivefunctions; they may not have
enhanced reproductivetness, nor resulted in the evolution of
mechanisms speci-cally designed to foster the development of MNs
(see sect.8). Rather, it is possible that MNs are constructed
bydomain-general processes of associative learning, and
arerecruited in the course of development to contributeto one or
more psychological functions, without eitherthe construction or the
recruitment processes havingbecome a specic target of gene-based
selection (Elmanet al. 1996).In this respect, MNs may be like beak
morphology in
Neotropical woodcreepers, which has been selected forforaging
and food manipulation (a non-social function, ana-logous to
visuomotor capability) but also has effects on songproduction (a
social function, analogous to action under-standing; Derryberry et
al. 2012). Another more closelyrelated example comes from
honeybees, which are ableto use associative learning to
discriminate among humanfaces (Dyer et al. 2005). Given the
taxonomic- anddomain-generality of associative learning, and the
factthat human faces were not part of the environment inwhich
honeybee nervous systems evolved, we can be surethat associative
learning is not an adaptation for face dis-crimination in
honeybees. However, when they are put inan environment where faces
are important, honeybeescan use associative learning about faces to
optimize theirforaging behavior. Another example, which may
beclosely related in a different way, is the area of thehuman
occipito-temporal cortex known as the visualword form area (VWFA;
Petersen et al. 1990). This areaplays an important role in reading,
but, given the recentemergence of literacy in human history, the
VWFA isvery unlikely to be a genetic adaptation for reading.Rather,
the reading-related properties of the VWFA areforged in the course
of development, by literacy training,from a system adapted for
generic object recognition.
3.4. Four kinds of evidence
Four evidence-based arguments are crucial in decidingbetween the
genetic and associative accounts. The rst pro-vided the inspiration
and foundation for the genetichypothesis. It suggests that the eld
properties of MNsindicate that they were designed for action
understanding.The terms design and purpose are used here as
theywere by G. C. Williams in his seminal work on Adaptationand
Natural Selection (Williams 1966). Williams describedadaptations as
designed by natural selection to fulll a par-ticular purpose, and
emphasized that the mark of an adap-tation is that it has features
making it peculiarly apt toachieve a specic end in a highly efcient
way. Forexample, An examination of the legs and feet of the
foxforces the conclusion that they are designed for runningand
walking, not for the packing or removal of snow (p.13). In a
similar way, supporters of the genetic hypothesisargue that
examination of the eld properties of MNs and, in particular, their
goal coding forces the con-clusion that MNs are designed for action
understanding.In section 3 we examine the eld properties of MNs
andsuggest that this argument is not compelling.
Cook et al.: Mirror neurons: From origin to function
182 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
The second argument is complementary to the rst; ithas a similar
form but advocates the associative hypothesis.It suggests that
research using conditioning proceduresshows associative learning to
be the right kind of learningto produce MNs. Specically, as we
discuss in section 5,the ways in which associative learning tracks
contingentrelationships, and enables contextual modulation of
theseconnections, makes it apt to produceMNs (and non-match-ing
visuomotor neurons) in typical developmentalenvironments.The second
pair of arguments draws on research examin-
ing the development of MNs and their modicationthrough
sensorimotor experience. Section 6 discussesresearch with infants
and adults that has been used tosupport a poverty of the stimulus
argument (Chomsky1975); to suggest that MNs emerge too early in
develop-ment or, more generally, after too little
sensorimotorexperience, to have been forged by associative
learning.In contrast, we offer a wealth of the stimulus
argument.Finally, section 7 focuses on evidence that, even in
adult-
hood, the properties of MNs can be changed in radical waysby
relatively brief periods of sensorimotor experience. Weargue,
against various objections, that this evidence issound and
therefore supports the associative hypothesisby showing that it has
produced novel, testable predictionswhich have been conrmed by
experiment.
4. Designed for action understanding
Supporters of the genetic hypothesis argue that examin-ation of
the eld properties of MNs shows that theyencode goals, and this
characteristic indicates that theywere designed by genetic
evolution to mediate actionunderstanding (Bonini & Ferrari
2011; Rizzolatti & Craigh-ero 2004; Rizzolatti & Sinigaglia
2010). We therefore beginour survey of the evidence by considering
how well theneurophysiological data accord with this view. The
termgoal affords numerous interpretations (Hickok 2009).We will
consider two denitions commonly adopted,assuming that MNs encode
goals if they encode object-directed actions (sect. 4.1) or
high-level action intentions(sect. 4.2).
4.1. Goals as object-directed actions
Early descriptions of MN eld properties reported thatpantomimed
actions (e.g., miming a precision grip in theabsence of an object)
and intransitive actions (e.g.,tongue-protrusion) did not elicit MN
responses (di Pelle-grino et al. 1992; Gallese et al. 1996). In
contrast, robustresponses were reported when monkeys observed
object-directed actions. This pattern raised the possibility
thatMNs encode goals in the sense that they are responsiveonly to
object-directed actions (di Pellegrino et al. 1992;Gallese et al.
1996).However, a close reading of the single-cell data suggests
that only a small subset of MNs appeared to have beendesigned
for encoding action goals in these terms. Asubset of the MNs
described in the early reports continuedto respond, albeit less
strongly, to pantomimed or intransi-tive actions (di Pellegrino et
al. 1992; Gallese et al. 1996,Figure 5b), and subsequent studies
conrmed that sizableproportions, perhaps the majority, of MNs
exhibit robust
responses to the observation of object-free body move-ment.
Kraskov et al. (2009) reported that 73% of MNresponses modulated by
observation of object-directedgrasping showed similar modulation
during observation ofpantomimed grasping. Also, substantial
proportions ofMNs respond to intransitive mouth movements such
aslip-smacking, lip-protrusion, and tongue-protrusion(Ferrari et
al. 2003).Single-unit data also show that, even when they are
responding to object-directed actions, MNs have eldproperties
suggesting that they were not tuned to do thisby genetic evolution.
For example, after training in whichtools were used to pass food
items to monkeys, MNswere discovered that respond to the
observation ofactions such as grasping with pliers (Ferrari et al.
2005).Similarly, audiovisual MNs respond to unnatural
soundsassociated with actions; for example, the sound of
metalstriking metal, plastic crumpling, and paper tearing(Keysers
et al. 2003; Kohler et al. 2002). Importantly,large numbers of
tool-use and audiovisual MNs respondmore to the sight of
tool-actions and to action soundsthan to the sight of gripping or
tearing executed with thehands. The fact that these MNs respond
maximally to unna-tural stimuli that is, stimuli to which the
evolutionaryancestors of contemporary monkeys could not
possiblyhave been exposed is hard to reconcile with the
genetichypothesis (Cook 2012; see sect. 7).
4.2. Goals as high-level intentions
The term goal has also been used to refer to what, at ahigh
level of generality, the actor intends to achievethrough their
behavior for example, grasp in order toeat (Fogassi et al. 2005) or
taking possession of anobject (M. J. Rochat et al. 2010).
Rizzolatti and Sinigaglia(2010, p. 269) state: only those [neurons]
that can encodethe goal of the motor behavior of another individual
withthe greatest degree of generality can be considered to
becrucial for action understanding. The suggestion thatMNs encode
high-level action intentions is made plausibleby reports that MN
responses to grasping can be modu-lated by the nal outcome of the
motor sequence (Boniniet al. 2010; Fogassi et al. 2005). It is also
consistent withreports that some broadly congruent MNs respond to
theobservation of multiple actions; for example, any graspingaction
executed with the hand or mouth (Gallese et al.1996).However, the
single-cell data again suggest that relatively
few MNs have the eld properties one would expect of asystem
designed by genetic evolution to represent high-level action
intentions. For example, Gallese et al. (1996)reported that during
action observation 37.5% of MNsresponded differently depending on
whether the actionwas executed with the left or right hand, and
64%showed direction sensitivity, preferring either left-to-rightor
right-to-left grasping actions. Similarly, many MNs(53%) respond
selectively to the observation of actions exe-cuted within
(peripersonalMNs) or beyond (extraperso-nal MNs), not the actors,
but the observing monkeysreach (Caggiano et al. 2009). The majority
(74%) of MNsalso exhibit view-dependent responses; some MNs
aretuned to egocentric (rst-person) presentation, whileothers
respond maximally to allocentric (third-person) per-spectives
(Caggiano et al. 2011). Each of these classes of
Cook et al.: Mirror neurons: From origin to function
BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2 183
-
MN is sensitive to features of action that fall well belowthe
greatest degree of generality of intentions such asgrasping in
order to eat or taking possession of anobject.The eld properties of
logically related MNs are perhaps
the hardest to reconcile with the idea that MNs weredesigned by
genetic evolution to mediate action under-standing by activating in
the observer the same goalthat is guiding the actors behavior.
Logically relatedMNs re when a monkey observes an action with
onegoal (e.g., placing food items on a table, with the intentionof
giving food to the monkey) and when the monkey exe-cutes an action
with a different goal (e.g., grasping thefood with a precision-grip
and bringing it to the mouth,with the intention of eating). Thus,
these MNs respondto different object-directed actions, with
different inten-tions, in observe and execute conditions.
5. The right kind of learning
The previous section argued that many MNs have eldproperties
incompatible with the hypothesis that theywere designed by genetic
evolution to mediate actionunderstanding via goal coding. In
complementary fashion,this section argues that research on the
roles of contingencyand contextual modulation in associative
learning enablesthe associative hypothesis to provide a unied
account ofall the MN eld properties reported to date.
5.1. Predictive relationships
It has long been recognized that associative learningdepends,
not only on contiguity events occurring closetogether in time but
also on contingency the degree towhich one event reliably predicts
the other. Where the pre-dictive relationship between two events is
weak that is,where one event is equally likely to occur in the
presenceand absence of the other event contiguous pairingsproduce
little or no learning (Elsner & Hommel 2004;Rescorla 1968;
Schultz & Dickinson 2000). The associativeaccount therefore
predicts that MNs will acquire sensori-motor matching properties
only when an individual experi-ences contingencies between sensory
events andperformed actions (Cooper et al. 2013b). This feature
ofassociative learning ensures that the matching propertiesof MNs
reect, not just chance co-occurrences, but sensor-imotor
relationships that occur reliably in the individualsenvironment.
Evidence that the human mirror mechanismis modied by contingent but
not by non-contingent sen-sorimotor experience has been reported by
Cook et al.(2010).Contingency sensitivity explains the mix of
strictly con-
gruent MNs, sensitive to the low-level features of
observedactions (type of grip, effector used, direction of
movement,viewpoint, proximity to the observer), and broadly
congru-ent MNs, responsive to multiple related actions
irrespec-tive of the manner of their execution. Both visual
andmotor systems are known to be organized hierarchically(Felleman
& Van Essen 1991; Giese & Poggio 2003; Jean-nerod 1994;
Perrett et al. 1989), comprising different popu-lations encoding
relatively low-level representations (e.g.,descriptions of
particular precision or power grips)and more abstract
representations (e.g., descriptions of
grasping). Therefore, contingencies can be experiencedbetween
both low- and high-level sensory and motor rep-resentations. When a
monkey observes itself performinga precision grip, the excitation
of sensory and motor popu-lations encoding a specic grip are
correlated. However,during group feeding, a monkey might observe
andperform a range of grasping actions, thereby causing corre-lated
excitation of higher-level visual and motoric descrip-tions of
grasping. Contingency sensitivity thereforeexplains the existence
of both strictly congruent MNs,tuned to a particular sensory
representation (e.g., a right-to-left precision grip executed with
the right hand viewedegocentrically), and broadly congruent MNs,
responsiveto the observation of a number of related
actions.Contingency sensitivity also explains the existence of
logically related, audiovisual, and tool-use MNs. Accordingto
the associative hypothesis, MNs acquire sensorimotorproperties
whenever individuals experience a contingencybetween seeing and
doing. There is no requirementthat contingencies are between action
performance andthe observation of the same action, or indeed of
naturalaction-related stimuli, such as the sight of animate
motionor sounds that could have been heard by ancestors of
con-temporary monkeys. Both monkeys and humans frequentlyexperience
non-matching sensorimotor contingencies,where the observation of
one action predicts the executionof another; for example, you
release and I grasp (Newman-Norlund et al. 2007; Tiedens &
Fragale 2003). The associ-ative account therefore explains in a
very straightforwardway why logically related MNs respond to
differentactions in observe and execute conditions. Equally,
theassociative account explains in a simple way why tool-use MNs
(Ferrari et al. 2005) develop when action per-formance is reliably
predicted by the sight of actionsperformed with tools (e.g., food
items being gripped withpliers), and why audiovisual MNs (Keysers
et al. 2003;Kohler et al. 2002) develop when action performance
pre-dicts characteristic action sounds (e.g., paper tearing
orplastic crumpling; Cook 2012).
5.2. Contextual modulation
Studies of conditioning indicate that learned responses areoften
subject to contextual control; if a stimulus is associ-ated with
two responses, each in a different context, thenthe context
determines which response is cued by thestimulus (Bouton 1993;
1994; Peck & Bouton 1990). Forexample, Peck and Bouton (1990)
initially placed rats in aconditioning chamber with a distinctive
scent (e.g.,coconut) where they learned to expect electric shock
fol-lowing a tone. The rats were then transferred to a
secondchamber with a different scent (e.g., aniseed) where thesame
tone predicted the delivery of food. The ratsquickly learned the
new contingency and conditioned fora-ging responses replaced
conditioned freezing. However,learning in the second phase was
context dependent.When the rats were returned to the rst chamber,
or trans-ferred to a third chamber with a novel scent, the tone
onceagain elicited freezing. By drawing on the components
ofassociative learning theory that explain this kind of effect,the
associative account of MNs can explain contextualmodulation of MN
ring (Cook et al. 2012a).Many, possibly all, of the ndings cited as
evidence that
monkey MNs code action goals can also be interpreted
Cook et al.: Mirror neurons: From origin to function
184 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
within an associative framework as evidence that MNs aresubject
to contextual control. For example, some MNsshow stronger visual
responses to object-directed graspingthan to pantomimed grasping in
object-absent contexts(Gallese et al. 1996), and in some cases, the
modulatinginuence of the object-context can be seen even whenthe
target object is occluded prior to contact with thehand (Umilt et
al. 2001). Similarly, MN responsesduring the observation of
grasping may be modulated bythe type of object being grasped
(Caggiano et al. 2012),with some MNs responding more strongly in
the presenceof high-value (food, or non-food objects predictive
ofreward), and some in the presence of low-value (non-food objects
not associated with reward) stimuli. In theclearest example, the
same motor act, grasping with a pre-cision grip, elicits different
MN responses dependent onwhether the action is observed in the
presence (grasp toplace) or absence (grasp to eat) of a plastic
cup(Bonini et al. 2010; Fogassi et al. 2005). Rather than
theplastic cup providing a cue to the actors intention, it mayact
as a contextual cue modulating the operation of twoassociations. In
the same way that the sound of the tone eli-cited different
behaviors when presented in the coconutand aniseed contexts (Peck
& Bouton 1990), observing aprecision grip may excite different
MNs in the cup-present and cup-absent contexts.Thus, many of the
eld properties cited as evidence of
goal (intention) coding by MNs can also be explained
bycontextual modulation within an associative framework.Under the
goal interpretation, these eld properties con-stitute direct
evidence that MNs mediate action under-standing. Under the
associative interpretation, they arevery interesting but not
decisive. The exibility apparentin the eld properties of MNs gives
them the potential tomake a useful contribution to social behavior.
However,further research, examining the behavior of whole
organ-isms, not only of neurons, is needed to nd out how
thispotential is realized (see sect. 9).
6. Wealth and poverty of the stimulus
Research involving infants (sect. 6.1) and adults (sect. 6.2)has
been used to support a poverty argument suggestingthat MNs emerge
too early in development or, more gen-erally, after too little
sensorimotor experience, to havebeen forged by associative
learning.
6.1. Mirroring in infancy
It has been argued that: (1) imitation is mediated by MNs(or a
mirror mechanism); (2) both human and monkeyinfants are able to
imitate observed actions when theyhave had minimal opportunity for
visuomotor learning;and (3) therefore, the associative account of
the origin ofMNs must be wrong (Gallese et al. 2011). The
structureof this argument is valid, but the evidence supporting
thesecond assumption (e.g., Heimann et al. 1989; Meltzoff&
Moore 1977; Nagy et al. 2005) has been challenged intwo respects.
Building on previous analyses (e.g., Anisfeld1996), a recent review
found evidence that human neo-nates copy only one action
tongue-protrusion and thatthis copying does not show the specicity
characteristic ofimitation or of MNs (Ray & Heyes 2011). Figure
2
illustrates the rst of these points. For each of the actiontypes
tested in young infants, it shows the number of pub-lished studies
reporting positive evidence of imitation andthe number reporting
negative evidence. This is a highlyconservative measure of how
often young infants havefailed imitation tests, because it is much
harder to publishnegative than positive results (Fanelli 2012).
Nonetheless,Figure 2 shows that the number of positive reports
subs-tantially exceeds the number of negative reports onlyfor
tongue-protrusion. Evidence that even the tongue-protrusion effect
lacks the specicity characteristic of imita-tion and MNs that it is
an exploratory response, ratherthan an effect in which action
observation is met with per-formance only of a similar action comes
from researchshowing that tongue-protrusion can be elicited by a
rangeof arousing stimuli, including ashing lights and livelymusic
(Jones 1996; 2006), and that it is greater wheninfants observe a
mechanical tongue or disembodiedmouth (Soussignan et al. 2011).2
More broadly, evidencethat the development of imitation is
crucially dependenton learning is provided by a study of 2-year-old
twinsshowing that individual differences in imitation were aresult
predominantly of environmental rather than geneticfactors (McEwen
et al. 2007), and by a recent study ofinfants indicating that
individual differences in associativelearning ability at 1-month
predicted imitative performanceeight months later (Reeb-Sutherland
et al. 2012).Turning from human to monkey infants, Ferrari et
al.
(2006) reported immediate imitation of tongue-protrusionand
lip-smacking in 3-days-old monkeys. However, theeffects were not
present on days 1, 7, and 14 postpartum,and it is not clear whether
they were replicated in a sub-sequent study using a similar
procedure (Paukner et al.2011). The later study did report
imitation of lip-smackingin monkeys less than one week old, but
this effect seems tohave been due to a low frequency of
lip-smacking in thecontrol condition, when infants were observing a
staticneutral face, rather than to an elevated frequency of
lip-smacking when the infants were observing
lip-smacking.Therefore, in common with the data from human
infants,studies of imitation in newborn monkeys do not
currentlysupport a poverty argument.
Figure 2. Summary of experiments seeking evidence of
gestureimitation in human infants (adapted from Ray & Heyes
2011).Gesture type refers to the target or modelled
movement.Positive frequencies (lighter bars) indicate the number
ofpublished experiments reporting positive cross-target
comparisons(i.e., infants performed the target action more often
afterobserving the target action than after observing an
alternativeaction). Negative frequencies (darker bars) indicate the
numberof experiments reporting failure to nd a signicant difference
incross-target comparison.
Cook et al.: Mirror neurons: From origin to function
BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2 185
-
A similar poverty argument suggests that the associativeaccount
must be wrong because (1) suppression of alphaband (613 Hz)
oscillations over central scalp locationsduring action observation
(and execution) reects the oper-ation of MNs or a mirror mechanism,
and (2) electroence-phalographic (EEG) studies indicate that both
human andmonkey infants show alpha suppression when they havehad
minimal opportunity for sensorimotor learning(Gallese et al. 2011;
Nystrm et al. 2011). In this case,both of the assumptions are
questionable. First, the func-tional signicance of alpha band
oscillatory activity ispoorly understood even in human adults, and
is yet moredifcult to interpret in human and monkey infants
where,for example, less information is available about the
source(Marshall & Meltzoff 2011). Second, human adult
studieshave traced the likely source of central alpha
suppressionduring action execution to the somatosensory cortex
(Hari& Salmelin 1997), suggesting that it may not index
motorprocessing at all.3 Third, even if alpha suppression doesindex
motor processing, it does not show that the motoractivation matches
or mirrors the actions observed (Mar-shall & Meltzoff 2011;
Pfurtscheller et al. 2000). Forexample, alpha suppression during
observation of lip-smacking, which has been reported in neonatal
monkeys(Ferrari et al. 2012), may reect a generalized readinessto
act, or arousal-related motor activation of tongue-protru-sion,
rather than motor activation of lip-smacking, andthereby the
occurrence of MN or mirror mechanismactivity. Furthermore, studies
of human infants, whichprovide superior source information, have
not shown thatcentral alpha suppression occurs when infants have
hadinsufcient correlated sensorimotor experience to build amirror
mechanism through associative learning. Indeed,studies of human
infants suggest an age-related trend con-sistent with the
associative hypothesis (see Marshall et al.[2011] for a
review).Sound evidence of MN activity in newborns which,
we suggest, has not been provided by research to date
onimitation and alpha suppression would be inconsistentwith the
associative model. However, it is important tonote that the
associative account is predicated on awealth of the stimulus
argument and therefore antici-pates MN activity in young infants
(Ray & Heyes 2011).This wealth argument points out that human
developmen-tal environments typically contain multiple sources of
thekind of correlated sensorimotor experience necessary tobuild
MNs; each of these sources is rich; and the mechan-isms of
associative learning can make swift and efcient useof these
sources. The range of sources available to younghuman infants
includes self-observation, being imitatedby adults, being rewarded
by adults for imitation, and thekind of experience in which, for
example, lip movementsmake the same smacking or popping sound when
observedand executed. Evidence of the richness of these
sourcescomes from studies showing that infants spend a high
pro-portion of their waking hours observing their own hands
inmotion (P. Rochat 1998; White et al. 1964); in
face-to-faceinteraction with a caregiver, they are imitated on
averageonce every minute (Jones 2009; Pawlby 1977; Uzgiriset al.
1989); and noisy actions, which provide an earlysource of acquired
equivalence experience, are amongthe rst that infants imitate
(Jones 2009). A common mis-conception about associative learning is
that it alwaysoccurs slowly. Directly relevant evidence that this
is not
the case comes from studies showing that, when the contin-gency
is high, infants can learn action-effect associations injust a few
trials (Paulus et al. 2012; Verschoor et al. 2010).
6.2. Motor training in adulthood
It has been claimed that the associative account cannotexplain
why motor experience obtained without visual feed-back can affect
perception of human biological motionrelated to that experience
(Gallese et al. 2011, p. 383).This claim assumes that the
perception of human biologicalmotion is mediated by MNs or a mirror
mechanism, andappeals to a subtle poverty argument; it suggests
that thefundamental properties of MNs the way in which theymap
observed with executed actions can be changed bymotor experience
alone, that is, in the absence of correlatedsensorimotor
experience.Two types of evidence, from studies that were not
designed to investigate MNs, have been cited in supportof this
subtle poverty argument (Gallese et al. 2011).First, when observing
point-light displays of whole bodymovements such as walking, from a
third party perspective,people are better able to recognize
themselves than torecognize their friends (Beardsworth &
Buckner 1981;Loula et al. 2005). Second, practice in executing
actionscan improve visual discrimination of those actions, evenwhen
actors are prevented from observing their move-ments during
execution (Casile & Giese 2006; Hechtet al. 2001). These motor
training effects, and the self-recognition advantage, are
interesting and importantphenomena in their own right. If they were
mediated bya mirror mechanism that is, a mechanism in which thereis
a direct, unmediated connection between visual andmotor
representations of action they would also supporta poverty
argument. However, a recent study provides evi-dence that these
effects depend on an indirect mechanismrepresenting temporal cues.
It shows, using avatar facialmotion stimuli, that the
self-recognition advantage is main-tained despite gross distortion
of the kind of spatial cuesthat characterize biological motion, but
is abolished byeven relatively minor disturbance of domain-general
tem-poral cues (Cook et al. 2012b). In the absence of appropri-ate
visual experience, actors appear to be able to use
theirconsiderable knowledge of the rhythmic characteristics oftheir
own actions to recognize and better represent allo-centric movement
displays. Thus, motor training effectsand the self-recognition
advantage are of independentinterest, but they do not support a
poverty argumentbecause current evidence suggests that they do
notdepend on a mirror mechanism.
7. Sensorimotor learning changes mirror neurons
7.1. Testing the predictions of the associative account
The associative account has been explicitly tested in
exper-iments examining the effects of laboratory-based
sensori-motor training on mirror mechanisms in human
adults.Building on the results of more naturalistic studies
(Calvo-Merino et al. 2005; 2006; Cross et al. 2006; Ferrari et
al.2005; Haslinger et al. 2005; Jackson et al. 2006; Keyserset al.
2003; Kohler et al. 2002; Margulis et al. 2009; Vogtet al. 2007),
these experiments have isolated the effects ofsensorimotor
experience from those of purely visual and
Cook et al.: Mirror neurons: From origin to function
186 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
purely motor experience. Using all of the measures ofmirror
mechanism activity commonly applied to humans(sect. 2.3), they have
shown that relatively brief periods ofsensorimotor experience can
enhance (Press et al. 2007;Wiggett et al. 2012), abolish (Cook et
al. 2010; 2012a; Gill-meister et al. 2008; Heyes et al. 2005;
Wiggett et al. 2011),reverse (Catmur et al. 2007; 2008; 2011), and
induce (Land-mann et al. 2011; Petroni et al. 2010; Press et al.
2012a)mirror mechanism activity. Each of these ndings conrmsa novel
prediction of the associative account: it reveals exi-bility of
exactly the kind one would expect if MNs/mechan-isms are forged by
sensorimotor associative learning. Incontrast, this kind of
exibility does not provide anysupport for the genetic hypothesis.
Indeed, if MNs were agenetic adaptation, some evolutionary
frameworks wouldpredict that the development of MNs would be
protectedor buffered against environmental perturbations thatcould
interfere with their adaptive function (Cosmides &Tooby 1994;
Pinker 1997). In the case of a genetic adap-tation for action
understanding, this would include pertur-bations with the potential
to divert MNs from codingproperties of action, rather than of
inanimate stimuli, andfrom coding similar, rather than dissimilar,
observed andexecuted actions.Evidence that MNs/mechanisms are not
resistant to
coding inanimate stimuli comes from studies showingthat
arbitrary sound, color, and shape stimuli can inducemirror MEP
(DAusilio et al. 2006; Petroni et al. 2010),fMRI (Cross et al.
2009; Landmann et al. 2011; Presset al. 2012a) and behavioral
effects (Press et al. 2007) fol-lowing sensorimotor training (Press
2011). For example,Press and colleagues gave participants
approximately 50minutes of sensorimotor training in which they
repeatedlyopened their hand when seeing a robotic pincer open,and
closed their hand when seeing the robotic pincerclose (Press et al.
2007). Prior to this training, the pincermovement elicited less
automatic imitation (see sect. 2.3)than human hand movement, but 24
hours after training,the automatic imitation effect was as strong
for the pincermovement as for the human hand.Evidence that
MNs/mechanisms are not resistant to
coding dissimilar actions comes from studies showing
thatnon-matching (or counter-mirror) sensorimotor trainingabolishes
automatic imitation (Cook et al. 2010; 2012a;Gillmeister et al.
2008; Heyes et al. 2005; Wiggett et al.2011), and reverses both
fMRI (Catmur et al. 2008) andMEP mirror responses (Catmur et al.
2007). Forexample, Catmur and colleagues gave participants
approxi-mately 90 minutes of non-matching sensorimotor trainingin
which they repeatedly made an index nger movementwhile observing a
little nger movement, and vice versa(Catmur et al. 2007). Before
this training the participantsshowed mirror MEP responses, for
example, observationof index nger movement elicited more activity
in anindex nger muscle than observation of little nger move-ment,
and vice versa for the little nger muscle. After train-ing, this
pattern was reversed, for example, observation ofindex nger
movement elicited more activity in the littlenger muscle than
observation of little nger movement.
7.2. Objections to sensorimotor training evidence
Objections to this evidence suggest, in various ways, that
itdoes not show that sensorimotor experience can change
MNs/mechanisms. For example, it has been suggestedthat the
evidence comes only from studies of object-freeactions and yet MNs
code only object-directed actions (Riz-zolatti & Sinigaglia
2010). However, a study of pianists hasshown that experience
modulates mirror responses toobject-directed actions (Haslinger et
al. 2005) and, as dis-cussed in sect. 4.1, monkey studies of
communicative ges-tures (Ferrari et al. 2003) and pantomimed
reachingmovements (Kraskov et al. 2009) have identied MNsthat code
object-free actions.A related concern is that, because they use
indirect
measures (fMRI, MEPs, and automatic imitation), ratherthan
single-cell recording, sensorimotor learning exper-iments may not
be measuring MN responses. Section 2.3reviewed the evidence for
human MNs from a range ofexperimental techniques. These include
conjunction ofneural responses during action observation and
perform-ance (Gazzola & Keysers 2009; Iacoboni et al. 1999;
Vogtet al. 2007), suppression of neural responses to cross-modally
(perceptual-motor or motor-perceptual) repeatedactions (Kilner et
al. 2009; Press et al. 2012b), muscle-specic MEPs (Catmur et al.
2011; Fadiga et al. 1995),and automatic imitation (Brass et al.
2001; Strmer et al.2000). In isolation, each of these measures is
imperfect(Caggiano et al. 2013), but together they provide
strongconverging evidence for human MNs. Sensorimotor learn-ing
effects have been demonstrated for all these measuresof mirror
responses: fMRI conjunction (Catmur et al. 2008;Landmann et al.
2011); repetition suppression (Press et al.2012a); MEPs (Catmur et
al. 2007; 2011; DAusilio et al.2006; Petroni et al. 2010); and
automatic imitation (Cooket al. 2010; 2012a; Gillmeister et al.
2008; Heyes et al.2005; Press et al. 2007; Wiggett et al. 2011).
Thus, conver-ging evidence using multiple techniques strongly
suggeststhat sensorimotor learning experiments are measuring and
changing MN responses. Furthermore, althoughexperiments specically
testing sensorimotor learning (inwhich sensory, motor, and
sensorimotor experience arecompared and/or controlled) have not
been performedusing single-unit recording, this conclusion is
supportedby single-unit data showing that experience with
toolscreates MN responses to observed tool use (Ferrari et al.2005;
M. J. Rochat et al. 2010; see sect. 4).Considerations regarding
anatomical specicity raise
another possible objection to the sensorimotor training
evi-dence: Sensorimotor experience may only affect neurons
innon-classical mirror areas (e.g., dorsal PMC). However,while
recordings of monkey MNs have mostly been con-ned to ventral PMC
and IPL, measurements in humansusing single-unit recording and fMRI
conjunction suggestthat MNs are more widespread (e.g., Arnstein et
al. 2011;Gazzola & Keysers 2009; Landmann et al. 2011;
Mukamelet al. 2010; Vogt et al. 2007). Furthermore, paired-pulseTMS
indicates that functional connections from dorsal (aswell as
ventral) PMC to primary motor cortex enhancemuscle-specic MEP
responses to action observation(Catmur et al. 2011). Thus, several
sources of evidencesuggest that MNs are not restricted to classical
mirrorareas. Therefore, even if sensorimotor experience
werealtering neuronal responses only outside ventral PMC
andinferior parietal cortex, it could still be affecting
MNs.However, there is also evidence that sensorimotor
learningaffects classical mirror areas. Many studies have
demon-strated effects of sensorimotor experience on classical
Cook et al.: Mirror neurons: From origin to function
BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2 187
-
mirror areas (Cross et al. 2009; Haslinger et al. 2005;
Land-mann et al. 2011; Margulis et al. 2009; Vogt et al.
2007);counter-mirror sensorimotor learning reverses ventralPMC and
inferior parietal cortex responses to observedactions (Catmur et
al. 2008); and such learning is supportedby ventral PMC-M1
connections (Catmur et al. 2011). Fur-thermore, localizing the
effects of sensorimotor learning tospecic neural populations using
repetition suppressionsuggested that sensorimotor experience
affects MNs in clas-sical mirror areas (Press et al. 2012a).
Therefore, it appearsthat MNs are not restricted to classical
mirror areas and thatsensorimotor experience has effects both on
classical mirrorareas and elsewhere.A nal possibility is that
counter-mirror training changes
relatively late neural responses to action observation,leaving
earlier responses, mediated by MNs, intact (Barch-iesi &
Cattaneo 2013). Such a nding might indicate thatcounter-mirror
responses result from a more indirectroute (e.g., via prefrontal
areas for rule retrieval) thanmirror responses. Barchiesi and
Cattaneo (2013) testedthis hypothesis using a task that is likely
to have provokedcoding of domain-general spatial cues, rather than
action-specic topographic cues, and therefore to have failed
toindex mirror responses at any time-point. A more recentstudy,
using a more specic test of mirror responses,found effects of
counter-mirror training on MEPs from200 msec, the earliest
time-point at which mirror responseshave been observed in monkeys
and humans (Cavallo et al.2013; see also Catmur et al. 2011).4
Thus, effects ofcounter-mirror training occur at the time when
complexinformation about the observed action has just reachedPMC,
making it improbable that mirror and counter-mirror effects occur
at different times. It is likely that a pre-frontal route is
involved during the training session, whenparticipants retrieve a
rule in order to implement taskinstructions (e.g., if index, do
little). However, thending that after counter-mirror training,
effects of trainingare present in MEPs from 200 msec suggests that
suchrule-based responding merely initiates associative
learning:after learning, action observation activates
counter-mirrorresponses as quickly as the original mirror
responses.In summary: Although there are currently no studies
sys-
tematically testing the effects of sensorimotor learning onMN
responses in monkeys, a substantial body of evidencefrom studies of
training and expertise in humans has con-rmed the predictions of
the associative account, showingthat mirror responses can be
changed in radical ways by sen-sorimotor learning. Furthermore,
these studies have pro-vided no evidence that MNs/mechanisms are
buffered orprotected against sensorimotor experience of a kind
thatmakes them code inanimate stimuli and dissimilar actions.
8. Other models: Canalization and exaptation
This article focuses on the genetic and associative accountsof
the origins of MNs because these were the rst modelsto be proposed,
and the associative hypothesis is themost fully developed
alternative to the standard, geneticview. For example, unlike other
alternatives, it has beenused to generate and test novel empirical
predictions.However, two other alternatives, which have been
motiv-ated in part by the data generated in these tests (seesect.
7) should be considered. One raises the possibility
that the development of MNs is canalized, and the otherthat it
represents an exaptation for action understanding.These are
interesting possibilities but, we argue, they arenot supported by
the evidence reviewed in sections 47.
8.1. Canalization
It has been suggested that MNs are acquired throughHebbian
learning (Keysers & Perrett 2004) and thattheir development is
supported or canalized by geneti-cally predisposed features of the
perceptual-motorsystem, including the tendency of infants to look
at theirown hands in motion (Del Giudice et al. 2009). On
onereading, this canalization hypothesis is identical in sub-stance
to the associative hypothesis; it is helpful in provid-ing a more
detailed neuronal model of how sensorimotorexperience makes MNs out
of motor neurons, and, in con-trast with the associative
hypothesis, it emphasizes self-observation over social interaction
as a source of relevantsensorimotor experience in development, but
otherwisethe canalization hypothesis is identical to the
associativeaccount. On this reading, the term Hebbian learning
isunderstood to be a synonym for associative learning,and the
canalization hypothesis suggests that if theinfants tendency to
look at their own hands in motion isan adaptation (Clifton et al.
1994; Meer et al. 1996) ifthis attentional bias evolved for
anything it was topromote the development of precise visuomotor
control,rather than MNs and action understanding.On another
reading, which we think is less likely to rep-
resent the authors intentions, Hebbian learning differsfrom
associative learning in depending on contiguityalone, rather than
both contiguity and contingency (seesect. 3.2), and the infant
preference for manual self-obser-vation evolved specically to
promote the development ofMNs and action understanding. If this
reading is correct,the canalization hypothesis is a hybrid of the
associative andgenetic accounts; it claims that MNs develop
through(Hebbian) sensorimotor learning and constitute a
geneticadaptation for action understanding. However, this
hybridmodel would not be supported by current evidence forthree
reasons. First, there is no evidence that the tendencyof infants to
look at their own hands evolved to promotethe development of MNs or
action understanding ratherthan visuomotor control (Del Giudice et
al. 2009). Second,experimental data and modeling work have
indicated thatthe sensorimotor learning which changes MNs depends
oncontingency as well as on contiguity (Cook et al. 2010;Cooper et
al. 2013b). Third, if MNs are forged by contin-gency-based
sensorimotor learning, there is no problem forevolution (or
scientists) to solve through canalization forMN development. If it
was based on contiguity alone, thereis a risk that sensorimotor
learning would produce lots ofjunk associations visuomotor neurons
mapping observedand executed actions that happen to have
co-occurred bychance. However, contingency-based (i.e.,
associative) learn-ing could produce the observed distribution of
strictly congru-ent, broadly congruent, and non-matching MNs all by
itself,without MN-specic canalization (see sect. 5.1).
8.2. Exaptation
Another interesting hybrid of the genetic and
associativehypotheses has been developed by Arbib and
colleagues
Cook et al.: Mirror neurons: From origin to function
188 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
(e.g., Arbib 2005; Oztop et al. 2006). They propose thatMNs are
produced, not by domain-general mechanismsof Hebbian or associative
sensorimotor learning, but bya special kind of sensorimotor
learning which receivesinput from self-observation of hand motion.
This specialkind of learning is an exaptation for action
understand-ing: It evolved from more domain-general mechanisms,such
as those producing canonical neurons, specicallyto promote action
understanding through the productionof MNs. This exaptation
hypothesis does not specify, inpsychological or neurological terms,
exactly what is dis-tinctive about the kind of sensorimotor
learning that pro-duces MNs. However, it suggests that some
extrastructure is required, both to constrain the variables
rel-evant for the system, and to track trajectories of
thoserelevant variables, and that the function of this
extrastructure is to ensure coding of goals or
hand-objectrelationships (Oztop et al. 2006, p. 269). Bonini
andFerrari (2011) recently advanced a similar exaptationhypothesis,
also motivated by the need to explain whyMNs consistently encode
goals. However, as we haveargued in section 4, the evidence from
single-unit record-ing in monkeys suggests that MNs do not
consistentlyencode goals. Therefore, the primary motivation
forinvoking exaptation is not compelling. Furthermore,there is no
evidence that the sensorimotor learninginvolved in MN development
is modied or constrainedrelative to the associative learning that
occurs in standardconditioning experiments. On the contrary, there
isexperimental evidence that it is sensitive to contingency,subject
to contextual modulation, and open to the encod-ing of both animate
and inanimate stimuli in exactly thesame way as standard
associative learning (see sects. 5and 7).A recent article (Casile
et al. 2011) adds another element
to the hybrid model advanced by Arbib and colleagues. Itsuggests
that a special, exapted form of sensorimotor learn-ing underwrites
the development of hand-relatedMNs, butthe development of facial
MNs is minimally dependent onexperience. This suggestion is
designed to accommodateevidence from studies of imitation and EEG
suppressionin newborns, which some authors have interpreted
asshowing that facial MNs are present at or shortly afterbirth. As
we reported in section 6, this evidence has beenchallenged on a
number of counts. Independent motivationand support for the idea
that hand and faceMNs have differ-ent origins would be provided by
evidence that faceMNs areless susceptible than hand MNs to
modication by sensori-motor experience. However, as far as we are
aware, thisnovel prediction of the hand/face hybrid model has
notbeen tested, and a recent study of improvement in facial
imi-tation suggests that face MNs are as susceptible to modi-cation
by sensorimotor experience as hand MNs (Cooket al. 2013). Thus,
until it is used to generate and testnovel predictions, the
hand/face hybrid model stands as anintriguing but essentially ad
hoc hypothesis.Hybrid modelling is a promising direction for
future
research. However, to preserve predictive power, it isessential
to check not only that hybrid models are consist-ent with existing
data, but also that they have indepen-dent support. We have argued
that both of theseconditions are met by the associative account,
and thatneither is currently fullled by canalization and
exaptationmodels.
9. A new approach to the function of mirrorneurons
We have argued that, at present, there is no positive evi-dence
that MNs are a genetic adaptation or exaptation, orthat their
development has been canalized, for actionunderstanding. However,
the associative hypothesis is func-tionally permissive; it does not
deny that MNs make a posi-tive possibly even an adaptive
contribution to socialcognition. Rather, the associative hypothesis
implies thata new approach is required to nd out what MNs
contrib-ute to social behavior.
9.1. From reection to theory-based experimentation
In the 20 years since MNs were discovered, theories relat-ing to
their function have been inspired by a method which(if you like a
pun) could be called reection. This methodfocuses on the eld
properties of the MNs found in asample of laboratory monkeys with
unreported develop-mental histories. It asks, usually without
reference to pre-existing computational or psychological theory,
whatneurons with these eld properties would be good for;that is,
what they might enable the animal to do. Forexample, early reports
that MNs discharged whenmonkeys saw and produced object-directed
actionsinspired the theory that MNs mediate action understand-ing
via motor resonance, when neither of these was anestablished
category of psychological functioning. Evennow, opposition to the
idea that MNs mediate actionunderstanding tends to be answered by
stressing theireld properties (Gallese et al. 2011). The
associativeaccount suggests that the reection method needs to
bechanged and extended in three principal ways.
9.1.1. Developmental history. If MNs were a geneticadaptation,
it is likely that their properties would be rela-tively invariant
across developmental environments. There-fore, it would be possible
to make valid inferences aboutspecies-typical properties of MNs
based on a relativelysmall and developmentally atypical sample of
individuals.If MNs are instead a product of associative learning,
thiskind of inference is not valid. Whether or not an individualhas
MNs, which actions are encoded by their MNs, and atwhat level of
abstraction, will all depend on the types ofsensorimotor experience
received by the individual in thecourse of their development.
Therefore, the associativeaccount implies that it is crucial for
studies of laboratorymonkeys to report, and ideally to control, the
animalsdevelopmental history; that is, the kinds of
sensorimotorexperience to which they have been exposed. It
alsosuggests that, if we want to know the species-typical
prop-erties of monkey MNs, it will be necessary to test monkeysthat
have received all and only the types of sensorimotorexperience
typically available to them under free-livingconditions. A
corollary of this is that we cannot assumethat the mirror
mechanisms found in the members of onehuman culture are
representative of the whole humanspecies. With its emphasis on the
role of social practices such as the imitation of infants by
adults, sports and dancetraining, and mirror self-observation in
driving the devel-opment of MNs, the associative account provides
specic,theory-driven motivation for cross-cultural studies
ofmirroring.
Cook et al.: Mirror neurons: From origin to function
BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2 189
-
9.1.2. System-level theory. If MNs were a genetic adap-tation,
one could argue that new categories of psychologicalfunctioning
such as action understanding and motorresonance are necessary to
characterize what they do.It could be argued that, since they were
speciallycreated by evolution, MNs are likely to have a highly
dis-tinctive, largely independent, and previously
unrecognizedpsychological function. In contrast, by showing that
estab-lished psychological theory associative learning theory can
cast light on the origin of MNs, the associativeaccount underlines
the value of embedding research onMN function within system-level
psychological and compu-tational theories of how the brain produces
behavior (Giese& Poggio 2003; Kilner 2011; Kilner et al.
2007a). Thisimplies that hypotheses about MN function shouldspecify
a part in a process a process that goes all theway from peripheral
sensory input to overt motor output that MNs are thought to fulll.
The name assigned to thispart is not important in itself. What is
important is that thehypothetical function of MNs is distinguished
clearly fromother components of the same overall process.
Forexample, in this kind of system-level, theory-guidedapproach,
action understanding would be distinguishedfrom components that are
likely to be more purely percep-tual (which might be called action
perception or actionrecognition), more purely motoric (e.g.,
actionexecution), or to constitute a higher level of understand-ing
(e.g., mentalizing). This approach would also make itclear whether
the hypothetical function is thought to beoptional or obligatory;
whether it can be, or must be,done by MNs. The kind of system-level
theoreticalapproach required in research on the functions of MNs
isexemplied by studies of their role in speech perception(Lotto et
al. 2009; Scott et al. 2009).A system-level theoretical approach
would also over-
come a problem that has haunted discussions of theaction
understanding hypothesis since MNs were discov-ered: Is this
hypothesis claiming that MN activity causes orconstitutes action
understanding? The former is anempirically testable hypothesis
suggesting that there is adistinctive behavioral competence (the
nature of whichhas not yet been specied, see sect. 3.1), called
actionunderstanding, to which the activity of MNs contributes.The
latter implies that the ring of MNs during actionobservation is, in
itself, a form of action understanding;it does not need to have
further consequences in order toqualify as action understanding.
This claim is notsubject to empirical evaluation; it is true, or
otherwise, byvirtue of the meanings of words.
9.1.3. Experimentation. Empirical (rather than constitu-tive)
claims about the function of MNs need to be testedby experiments
looking for, at minimum, covariationbetween MN activity and
behavioral competence, and,ideally, testing for effects on
behavioral competence ofinterventions that change MN activity. A
brief survey ofrecent research of this kind using fMRI, TMS, and
theeffects of focal brain lesions in human participants is
pro-vided in the next section. At present, this research faces
twomajor challenges. First, because the hypothetical functionsof
MNs typically are not dened in the context of a system-level
theory, it is difcult to design appropriate controltasks. For
example, if an experiment is testing the hypoth-esis that MNs play
a causal role in action understanding,
should it control for the possibility that they instead playsome
role in action perception? If so, what kind of behav-ioral
competence is indicative of action perception ratherthan action
understanding?5 To date, only a smallnumber of studies (e.g.,
Pobric & Hamilton 2006)include control conditions designed to
address this issue.The second major challenge is that, with rare
exceptions
(Mukamel et al. 2010), MN activity cannot be localized
pre-cisely within the human brain. Consequently, many studiesassume
that activity in the ventral PMC and IPL areashomologous to those
in which MNs have been found inmonkeys is MN activity, and that
behavioral changesbrought about through interference with the
functioningof these areas are due to interference with MNs.
Theresults of such studies are of interest regardless ofwhether
they relate to MNs. However, it is unsatisfactoryto assume that
they relate to MNs, because, in monkeys,it is likely that fewer
than 20% of the neurons in these clas-sical mirror areas are
actually MNs, and because there isevidence of MNs in non-classical
areas in both monkeysand humans (see sect. 2.1). Techniques such as
fMRI rep-etition suppression and TMS adaptation (Cattaneo et
al.2011; Silvanto et al. 2007) hold some promise as meansof
overcoming the localization problem with human partici-pants, by
isolating behavioral effects to specic populationsof neurons.
Guided by system-level theory, future studiescould use these
techniques with a range of tasks to isolateprocesses in which MNs
are involved.Alongside the development of techniques such as
fMRI
repetition suppression and TMS adaptation for use withhuman
participants, it would be valuable to conductanimal studies that,
not only document the eld propertiesof MNs, but also examine how
those properties relate tobehavioral competence. For example, are
animals withMNs for actions X and Y better than other animals of
thesame species at behavioral discrimination of X and Y, orat
imitating X and Y? Studies of this kind have been dis-missed as
impractical on the assumption that they wouldhave to involve
monkeys, which are demanding and expens-ive laboratory animals, and
that between-group variation inMN activity would have to be induced
via lesions or disrup-tive TMS. However, the associative account
suggests that,in the long term, it may be possible to overcome
these pro-blems by establishing a rodent model and using
sensorimo-tor training to induce between-group variation in
thenumber and type of MNs present in rodent brains. Ifthe
associative account is correct, rodents are likely to havethe
potential to develop MNs because they are capable ofassociative
learning. Whether or not they receive in thecourse of typical
development the sensorimotor experiencenecessary to realize this
potential, it could be provided byvarious regimes of
laboratory-based sensorimotor training.
9.2. Early signs
Given the theoretical limitations and methodological chal-lenges
faced by research to date on the functions ofMNs, it is very
difcult indeed to form a consistent andpotentially reliable picture
of where their ndings arepointing. Nonetheless, for completeness,
we offer the fol-lowing brief overview of research relating to the
two mostcommonly investigated hypotheses MN activity contrib-utes
to action understanding and to imitation. We rst sum-marize the
results of meta-analyses of functional imaging
Cook et al.: Mirror neurons: From origin to function
190 BEHAVIORAL AND BRAIN SCIENCES (2014) 37:2
-
data, which, by identifying commonalities across
multiplestudies, highlight candidate brain areas involved in
thesehypothesized MN functions. Building on the fMRIstudies,
patient data and studies using disruptive TMSapplied to classical
mirror areas have the potential to ident-ify causal, rather than
purely correlational, relationshipsbetween MN activity and
behavioral competence.
9.2.1. Action understanding. As discussed in section
3.1,consensus regarding the term action understanding hasyet to be
reached. In terms of the involvement of mirrorareas in perceiving
others actions, a recent meta-analysisof fMRI data on action
observation revealed that obser-vation of hand movements produces
respons