NeuroImage xxx (2008) xxx–xxx

YNIMG-05487; No. of pages: 7; 4C:

Contents lists available at ScienceDirect

NeuroImage

j ourna l homepage: www.e lsev ie r.com/ locate /yn img

ARTICLE IN PRESS

I learned from what you did: Retrieving visuomotor associationslearned by observation

Elisabetta Monfardini, Andrea Brovelli, Driss Boussaoud, Sylvain Takerkart, Bruno Wicker ⁎CNRS and Aix-Marseille University UMR 6193, Mediterranean Institute for Cognitive Neuroscience 31 chemin Joseph Aiguier, 13402, Marseille, France

⁎ Corresponding author.E-mail address: wicker@incm.cnrs-mrs.fr (B. Wicker

1053-8119/$ – see front matter © 2008 Elsevier Inc. Alldoi:10.1016/j.neuroimage.2008.05.043

Please cite this article as: Monfardini, E., eNeuroImage (2008), doi:10.1016/j.neuroima

a b s t r a c t

a r t i c l e i n f o

Article history:

Observational learning allo Received 14 January 2008Revised 15 May 2008Accepted 19 May 2008Available online xxxx

ws individuals to acquire knowledge without incurring in the costs and risks ofdiscovering and testing. The neural mechanisms mediating the retrieval of rules learned by observation arecurrently unknown. To explore this fundamental cognitive ability, we compared the brain responses whenretrieving visuomotor associations learned either by observation or by individual learning. To do so, we askedeleven adults to learn two sets of arbitrary visuomotor associations: one set was learned through theobservation of an expert actor while the other was learned by trial and error. During fMRI scanning, subjectswere requested to retrieve the visuomotor associations previously learned under the two modalities. Theconjunction analysis between the two learning conditions revealed a common brain network that includedthe ventral and dorsal lateral prefrontal cortices, the superior parietal lobe and the pre-SMA. This suggeststhe existence of a mirror-like system responsible for the storage of rules learned either by trial and error or byobservation of others' actions. In addition, the pars triangularis in the right prefrontal cortex (BA45) wasfound to be selective for rules learned by observation. This suggests a preferential role of this area in thestorage of rules learned in a social context.

© 2008 Elsevier Inc. All rights reserved.

Introduction

In daily life many arbitrarily fixed and generally adopted“rules” are learned and retrieved, allowing us and otheranimals to anticipate relevant events and adapt to a novelcontext. These rules are often associations of sensoryenvironmental events (e.g. image, symbol, context, etc.) withparticular motor responses or actions. If the relation betweenthe visual stimulus, the action, and its outcome is arbitraryand causal, we refer to it as arbitrary visuomotor learning(Wise et al., 1996; Wise and Murray, 2000). If one excludeslearning through verbal explanations and instructions, whichnecessitates language, learning by trial and error and learningby observation are two mechanisms of particular interest forthe development of culture in human infants and adults(Castro and Toro, 2004; Cavalli-Sforza and Feldman,1981). Themajor difference between these two mechanisms is that theaction and its consequence are experienced by the learnerhimself or herself in the case of learning by trial and error, orby another individual in the case of learning by observation. In

).

rights reserved.

t al., I learned from what yge.2008.05.043

fact, learning by observation allows individuals to acquireuseful knowledge without incurring in the costs and risks ofdiscovering and testing (Bandura et al., 1977; Boyd andRicherson, 1985, 1988). As Bandura (1977) stated it: “Learningwould be exceedingly laborious, not to mention hazardous, ifpeople had to rely solely on the effects of their own actions toinform them what to do. Fortunately, most human behaviouris learned observationally through modelling: from observingothers one forms an idea of how new behaviours areperformed, and on later occasions this coded informationserves as a guide for action.” (Bandura et al., 1977).

Several neuroimaging studies have established that learn-ing arbitrary visuomotor associations by trial and errorengages a large brain network including the frontal–parietalsystem, the basal ganglia and medial temporal structures(Deiber et al., 1997; Toni and Passingham, 1999; Toni et al.,2001; Eliassen et al., 2003; Law et al., 2005; Brovelli et al., inpress). In parallel, various studies have shown that theprefrontal cortex is particularly involved in the storage andretrieval from long-term memory of known rules, or pre-scribed guides for action (Bunge et al., 2003; Bunge, 2004;Crone et al., 2006; Donohue et al., 2005). However, despite itsoutstanding scientific interest, the neural bases of learningand retrieval of visuomotor associations when learned byobservation remain surprisingly unexplored.

ou did: Retrieving visuomotor associations learned by observation,

2 E. Monfardini et al. / NeuroImage xxx (2008) xxx–xxx

ARTICLE IN PRESS

In the present study, we investigate the neural representa-tions of the retrieval of visuomotor associations learned bytrial and error or by observation. Retrieval is defined as therecovery from memory of the correct action associated to agiven visual stimulus. In other words, we indirectly testwhether observational learning is mediated by specific neuralcircuits linking environmental information such as visualstimuli with the reenactment of others' actions to generateinternal representations of arbitrary associations. To do so,eleven adult subjects were asked to learn, prior to the fMRIscanning session, two sets of visuomotor associations: one setwas learned through the observation of an expert actor whilethe other was learned by trial and error. To explore the neuralcorrelates of learning by observation, we compared the brainactivations of subjects during the retrieval of associationslearned in the two modalities.

Materials and methods

Participants

Elevenhealthy, right-handedvolunteers (7males, 4 females)participated in the study (mean age: 26.3±4.2 years). Allsubjects were screened to rule out medication use, history ofneurological or psychiatric disorders, head trauma, substanceabuse, or other serious medical conditions. Written consentwas obtained after the procedure had been fully explained. Thestudy was approved by the local ethics committee and wasconducted in accordance with the Declaration of Helsinki.Volunteers were paid for their participation.

Fig. 1. (A) Learning session. Subjects were asked to associate abstract visual stimuli with joysthe joystick up, down, right or left. A visual feedback stimuli indicated if themovement was clearned one set of associations by trial and errors and another set by observation of an expertintermixed pseudo-randomly. In all conditions, the visual stimuli was presented for 1.5 s anexecuting a joystickmovement. After the movie, the subjects were asked to judge if the actorto “yes” or “no” displayed on the right or left side of the screen. Actor's movement was 50%colour in this figure legend, the reader is referred to the web version of this article.)

Please cite this article as: Monfardini, E., et al., I learned from what yNeuroImage (2008), doi:10.1016/j.neuroimage.2008.05.043

Design and experimental conditions

Learning sessions prior to scanningPrior to fMRI scanning, subjects learned arbitrary associa-

tions between abstract visual stimuli (linear segments com-bined to formwhite shapes on a black background—Fig. 1) andmotor responses (joystick movements). Each subject had tolearn one set of three visuomotor associations by trial and errorand another set by observing an expert actor performing thevisuomotor associations in front of him or her. The order of thetwo learning sessions was randomized between subjects.

During the trial and error learning session, visual stimuliwere presented on a computer screen in front of the subjects.Motor responses were recorded with a joystick. Afterpresentation of the visual stimulus (1.5 s), the subject had tochoose one of four possible joystick movements (up, down,right or left). The visual feedback (a green/happy or red/sadsmiley-like face) then indicated whether the movement wascorrect or incorrect. The correct stimulus–response associa-tions were kept constant throughout the experiment. Toensure that participants had learned the associations, thesession ended when the subject had given 4 consecutivecorrect responses to each of the 3 visual stimuli. A pilot studywas conducted on 5 subjects in order to validate the learningcriterion. A mean of 46.36 trials was necessary to reach thecriterion.

During learning byobservation, subjects had to learn throughthe observation of an expert individual demonstrating thecorrect visuomotorassociations. Subjectswere seatedbeside thedemonstrator and close enough to observe t he visual stimuli

tick movements. After presentation of a visual stimulus (1.5 s), the subjects had to moveorrect (happy green face: example 1), or incorrect (sad red face: example 2). The subjectsactor executing the task. (B) Scanning session. Trials from three different conditions wered followed by a blank screen of variable duration and a short video of an actor’s handmade a correct movement or not by pressing a right or left mouse button correspondingright and 50% wrong. ISI, interstimulus interval. (For interpretation of the references to

ou did: Retrieving visuomotor associations learned by observation,

3E. Monfardini et al. / NeuroImage xxx (2008) xxx–xxx

ARTICLE IN PRESS

on the computer screen and the joystick movement (distance=approximately 50 cm). A pilot study was conducted on 5subjects, in order to estimate the approximate number of trialsrepetition needed to learn the associations. To ensure that theparticipants had learned the visuomotor associations, thesession ended when the subjects told the experimenter theyknew the associations. The behavioral performances obtainedduring fMRI scanning were used to control that learning byobservation had occurred prior to scanning (cf. SupplementaryTable 1). A mean of 15.27 trials was necessary to learn theassociations.

Scanning session

Stimuli and experimental design. Subjects were scannedapproximately 30 min after the learning sessions using anevent-related fMRI paradigm with three trial types (condi-tions). Each trial started with a fixation cross (1 s), immedi-ately followed by the presentation of the visual stimulus(1.5 s). In the trial and error (TE) condition, the correct actionassociated to the presented visual stimulus had been learnedby trial and error previous to scanning. In the learning byobservation (LeO) condition, the stimuli had been previouslylearned by observation. A third trial type served as controlcondition (CONT), in which arrows pointing in one of fourdirections were used as visual stimuli (Fig. 1). The stimulusproperties in the control condition were very similar to thosein the TE and LeO condition. Subjects were not informed thatthe arrow indicated the correct direction of movement. For allconditions, the stimulus presentation was followed by a videosequence of 3 s showing the hand of an actor executing ajoystick movement. A variable delay between 5 and 12 s wasintroduced between the stimulus and the video presentationin order to dissociate the BOLD responses produced by the twoevents. After each movie, the subjects had to judge whetherthe joystick movement executed by the actor was correct ornot. Subjects responses were collected using a two-buttoncomputer mouse; the button associated to the “correct”response was randomly assigned across trials. The actor'sresponses were correct in 50% of the trials. Subjects performeda total of 40 trials of each of the three conditions, presentedpseudo-randomly in four blocks. Each block consisted of 30trials (10 of each condition), and the presentation order wasrandomized between subjects. Behavioral data (accuracy andreaction time) were collected during the scanning sessions.The stimuli were projected onto a screen positioned at theback of the scanner. Subjects could see the video reflected in amirror (15×9 cm) suspended 10 cm in front of their face andsubtending visual angles of 42° horizontally and 32° vertically.

Image acquisition. Images were acquired using a 3-T whole-body imager. For each participant, we first acquired a high-resolution structural T1-weighted anatomical image (inver-sion–recovery sequence, 1×0.75×1.22 mm) parallel to theAC–PC plane, covering the whole brain. For functionalimaging, we used a T2*-weighted echo-planar sequence at 36interleaved 3.5-mm-thick axial slices with 1 mm gap (TR=2400 ms, TE=35 ms, flip angle=80°, FOV=19.2×19.2 cm,64×64 matrix of 3×3 mm voxels).

Data processing and statistical analysis

Image processing and analysis of fMRI data were conductedwith SPM2software (http://www.fil.ion.ucl.ac.uk/spm/software

Please cite this article as: Monfardini, E., et al., I learned from what yNeuroImage (2008), doi:10.1016/j.neuroimage.2008.05.043

/spm2/). The first five volumes of each participant's data werediscarded to allow for longitudinal relaxation timeequilibration.Functional images for each subject were slice-time corrected toa slice acquired halfway through image acquisition in order tocorrect for temporal differences (up to 2.4 s) between slicesacquired early, and those acquired late in the image volume. Allvolumes were realigned to the first volume to correct for headmovement between scans. A mean image was created usingthe realigned volumes. The mean image was spatially normal-ized to the standard EPI template given in the SPM software.All images were then spatially normalized using the norma-lization parameters determined during the normalization ofmean image to EPI template. Datawere then smoothed using an8 mm full-width-at-half-maximum isotropic Gaussian kernelto accommodate inter-subject differences in anatomy. Finally,the data were normalized and high-pass filtered (128 s).

The statistical analysis of the pre-processed BOLD signalswas performed using a generalized linear model (GLM)approach. To dissociate two events per trial (the stimulus andvideo presentation), two regressors were constructed per trialtype by convolving the canonical haemodynamic responsefunction (HRF)withdelta functions aligned either on the timeofstimulus or video presentation. The design matrix contained atotal of six regressors (two regressors for each of the threeconditions). Since we hypothesized that association retrievaloccurs at stimulus presentation, we aligned our events ofinterest to the onset of the stimulus.

The regression coefficients (the beta values) were esti-mated for each subject, and were then taken to the random-effects level. All the fMRI statistics and p values arise fromgroup random-effects analyses. We considered as activatedbrain regions those clusters of more than 10 contiguous voxelswith pb0.001 at the voxel level (uncorrected for multiplecomparisons).

Results

Eleven subjects participated in the fMRI study. Prior to thescanning session, subjects learned two sets of three arbitraryvisuomotor associations. One set was learned by trial and errorand one set by observation of an expert actor performing thetask. During the scanning session, the subjects were tested inthree different conditions: i) retrieval of associations learnedby trial and error; ii) retrieval of rules learned by observation;and iii) a control condition using arrows as stimuli (see me-thods, Fig. 1).

The mean percentage of correct responses for each exper-imental condition acquired during fMRI scanning confirmedthat subjects could perfectly retrieve the visuomotor asso-ciation for all stimuli (Supplementary Table 1). There was nosignificant difference across conditions (TE, LeO and CONT)in reaction times (F(2,768)=0.238, pN0.05), suggesting thattask difficulty was comparable in all experimental conditions(Supplementary Table 2).

We compared the brain activations associated with theperception of three types of stimuli in the trial and error (TE),learning by observation (LeO) and control (CONT) conditions.Since we hypothesized that the retrieval of the correctassociation occurs at stimulus presentation, we searched forsignificant differences in brain activity by comparing theregressors aligned on the stimulus. To reduce unspecific effectsrelated to the retrieval of visuomotor associations not learnedeither by trial and error or by observation (i.e., the arrows), we

ou did: Retrieving visuomotor associations learned by observation,

Fig. 2. (A) Brain regions that were more active while subjects viewed stimuli they hadrecently learned by trial and error or by observation than while viewing control stimuli(arrows; pb0.01 FDR corrected, cluster size N10 functional voxels). (B) Plots of the meanvalue of the parameter estimates (arbitrary units, mean centered) for the maxima of theventral and dorsal pars opercularis (vPO, dPO). a, TE; b, LeO; c, CONT; d–f, movieconditions.

4 E. Monfardini et al. / NeuroImage xxx (2008) xxx–xxx

ARTICLE IN PRESS

subtracted the brain activity in the control condition (CONT)from the TE and LeO conditions. We then performed aconjunction analysis between the contrasts ‘TE’–‘CONT’ and‘LeO’–‘CONT’ to explore the brain areas commonly activatedboth in the LeO and TE conditions. This analysis revealed BOLDsignal increases bilaterally in three distinct foci of the lateralprefrontal cortex (BA46; ventral pars opercularis BA44; dorsalpars opercularis BA44/45). This network spread from theinferior to the superior posterior parietal cortex bilaterally, inthe dorsal posterior cingulate gyrus extending dorsally to thepre-SMA, in the posterior ventro-temporal visual areas, and inthe thalamus (Fig. 2; Table 1A). This network includes areasthat activate during the retrieval of associations recentlylearned either by trial and error or by observation. To assesswhich brain areas are specifically involved in rule retrievalafter observational learning, we performed a direct compar-ison between conditions LeO and TE. Significant activationswere found in the right prefrontal cortex (pars triangularis,BA45), in the inferior parietal lobule (BA40) and in the occipitalprimary visual areas (BA17/18; Fig. 3 and Table 1B).

Discussion

Although various aspects of rule retrieval and their neuralcorrelates have been explored in recent studies (Bunge et al.,2003; Crone et al., 2006; Donohue et al., 2005), none haveaddressed whether these cerebral areas can be influenced bythe type of preceding learning. In this study, we manipulatedhow visuomotor associations were learned prior to scanningand investigated their brain correlates during retrieval.

We compared the brain activations triggered by thestimulus presentation, that is, when subjects retrieve frommemory the correct action associated to the stimulus. Weused abstract visual stimuli to minimize verbalization (inter-nal speech) effects that could account for differences in brainactivations between the TE/LeO conditions and the CONTcondition. In addition, the training sessions prior to scanningwere matched to ensure that subjects performed the taskduring scanning without errors. Indeed, the analysis ofbehavioral data confirmed that the level of performance andreaction times were statistically identical in all conditions. Inorder to control for effects not deriving from the type oflearning, we used a control condition (CONT) where the visualstimuli were arrows pointing in the 4 joystick directions. Thiscontrol condition was subtracted from the TE and LeO con-ditions. Therefore, we argue that differences in brain activa-tions between conditions TE and LeO are exclusively related tothe type of learning.

Common networks involved in retrieval of associations learnedby trial and error and by observation

The results from the conjunction analysis confirmed theinvolvement of a brain network composed of the rightventrolateral and anterior prefrontal cortices, pre-SMA, andparietal cortex when retrieving newly acquired rules (Bungeet al., 2003; Bunge, 2004; Crone et al., 2006; Donohue et al.,2005). This is consistent with a role of the inferior frontaljunction in the retrieval of actions associated with symboliccues (Donohue et al., 2005). It also confirms that the posteriormedial temporal gyrus plays a general role in storing actionknowledge, more specifically in representing arbitrary asso-ciations between symbols and associated rules for how to act

Please cite this article as: Monfardini, E., et al., I learned from what yNeuroImage (2008), doi:10.1016/j.neuroimage.2008.05.043

(Donohue et al., 2005). Our results show that the samenetwork is recruited also during the retrieval of appropriateaction representations that have been learned via theobservation of others' behaviors. A straightforward interpre-tation is that during both types of learning, the brain builds aninternal model linking an executed or observed action with avisual stimulus and a feedback. The generation of such inter-nal models necessitates the activation of brain areas coding foraction execution and observation. Except for themost anteriorfrontal area (BA46), the activations in the dorsal and ventralpart of pars opercularis and in the parietal cortex have beenconsistently reported in studies where subjects execute,observe, imitate or understand actions, intentions or emo-tions, pretend to use tools, or generate verbs (Buccino et al.,2001; Gazzola et al., 2006, 2007; Grèzes et al., 2003; Iacoboniet al., 2005; Jeannerod, 2001; Rizzolatti et al., 1996; 2001;Wicker et al., 2003a). These brain areas are thought to be corecomponents of the mirror neuron system (MNS) in humansand may mediate action and intention understanding mecha-nisms (Rizzolatti and Craighero, 2004). In the present study,Fig. 2B illustrates that the prefrontal and parietal regionsactivate during stimulus perception and association retrieval,and also during observation of movies (i.e., observation of

ou did: Retrieving visuomotor associations learned by observation,

Table 1

Anatomical regions BA MNI coordinates

x y z PeakZ score

AR superior parietal lobe 7 24 −72 52 5.42L superior parietal lobe 7 −16 −78 52 5.13R inferior parietal lobe 40 48 −34 50 5.00L inferior parietal lobe 40 −34 −50 50 3.56R anterior cingulate gyrus(also pre-SMA)

24 4 16 44 5.15

R posterior cingulate gyrus 31 10 −46 18 3.51R inferior temporal gyrus 19 52 −60 −12 4.95L fusiform gyrus 37 −28 −74 −24 4.81L fusiform gyrus 37 −30 −36 −32 4.13R pulvinar 10 −26 12 4.74L middle frontal gyrus 46 −42 48 12 4.70R middle frontal gyrus 46 48 42 14 4.57L middle frontal gyrus 6 −22 0 50 4.40R inferior frontal gyrus 44/45 54 14 24 4.06L inferior frontal gyrus 44/45 −46 14 36 3.56R inferior frontal gyrus 44 38 20 0 4.13L inferior frontal gyrus 44 −36 14 −6 3.87L inferior frontal gyrus 44 −50 4 42 3.58R middle occipital gyrus 19 28 −98 −8 4.54L middle occipital gyrus 19 −26 96 −6 4.05Lingual gyrus 18 0 −72 −16 3.48R hippocampal gyrus 20 8 −18 3.90R calcarine sulcus 2 −64 0 3.75

BR cuneus 17/18 10 −104 18 5.18R inferior frontal gyrus (GFi) 45 42 20 18 3.91R inferior parietal lobule 40 60 −24 30 3.88R caudate nucleus 12 18 4 3.58

(A) Coordinates and peak Z value of functional activations common to LeO and TEconditions (FDR corrected pb0.01). (B) Location and peak Z value of functionalactivations found in LeO vs TE (p not corrected, cluster size N10 functional voxels). R,right hemisphere; L, left hemisphere.

Fig. 3. (A) Right lateral prefrontal region exhibiting learning by observation effects(Talairach coordinates: 42 20 18). (B) Percent change in bold signals observed in thisregion for TE (blue), LeO (green), CONT (red) and movies (light blue). Error bars indicatestandard error. The right lateral PFC shows a significantly stronger BOLD responseduring movies and LeO conditions than during TE (film–TE, p=0.006; LeO–TE, p=0.01).No significant difference was found between LeO and movies conditions (p=0.6). (Forinterpretation of the references to colour in this figure legend, the reader is referred tothe web version of this article.)

5E. Monfardini et al. / NeuroImage xxx (2008) xxx–xxx

ARTICLE IN PRESS

joystick movements). This demonstrates their role in actionobservation. Although we did not have an experimentalcondition to map the brain areas involved in action execution,our results suggest that a subset of the mirror systemmediates the retrieval of visuomotor associations recentlylearned by trial and error or by observation. Subject's ownactions and observed actor's actions are stored in the same‘mirror’ areas, suggesting the presence of a mirror-like systemthat stores the representations of recently learned visuomotorassociations. This interpretation of our findings implies that,in addition to the classically describedmirror neurons that fireduring the execution and observation of the same or broadlyidentical motor act (e.g. a joystick movement), a subset ofthese neurons might be visually triggered by the presentationof an abstract stimulus that has been previously associated to aspecific action. The ability to store visuomotor associationslearned by observation might rely on a kind of mirror activityof prefrontal areas, more precisely in the ventral and dorsalpart of the pars opercularis. Similarly to the understanding ofsensations (Keysers et al. 2004), emotions (Wicker et al.,2003b) and pain (Singer et al., 2004), observational learningmight be supported by a reenactment of the experience of themodel in the observer. Albeit theoretically (Gallese andGoldman, 1998), literature in the field of rule retrieval andlearning has not been linked to literature on mirror neuronsystem. Herewe provide suggestive evidence of a possible roleof the mirror system in observational learning of visuomotorassociations.

Please cite this article as: Monfardini, E., et al., I learned from what yNeuroImage (2008), doi:10.1016/j.neuroimage.2008.05.043

A specific network for the retrieval of associations learned byobservation

The observation of abstract stimuli and their association tospecific movements produced activations in a network of brainareas comprising the right pars triangularis (BA 45), the rightinferior parietal lobule and the posterior visual areas. To betterunderstand the role of the right pars triangularis (BA45), weperformed a Student's t-test on the average percent change inBOLD signal in the right pars triangularis during retrieval andmovie observation. The analysis revealed similar BOLDresponses when subjects retrieved the motor action associatedwith a visual stimulus in the LeO condition and when subjectswatched a video sequence of a hand performing a joystickmovement (Fig. 3). This suggests that the pars triangularis isengaged both during observation of action and during theretrieval of the correct movement previously associated with astimulus by observation of another's action. This is consistentwith recent data showing that the right pars triangularis isinvolved in action observation, but not in imitation, hence not inexecution (Molnar-Szakacs et al., 2005). Because of theseproperties, authors have suggested that this region should notbe considered as part of themirror system, but rather as relatedto the suppression of movement execution during actionobservation and motor imagery (Deiber et al., 1998; Molnar-Szakacs et al., 2006). Our results thus suggest that the mereperception of an abstract stimulus previously associated with aspecific movement is sufficient to activate this area and hencethat this region plays an important role in storing informationabout other's actions performed in a given context. One couldfurther propose that this area monitors the flow of informationwithin the fronto-parietal network, to store the information as

ou did: Retrieving visuomotor associations learned by observation,

6 E. Monfardini et al. / NeuroImage xxx (2008) xxx–xxx

ARTICLE IN PRESS

an internal representation about the actions of the demonstra-tor in relation with the visual stimulus and the feedback.Interestingly, the rostral part of the IPS has been recentlyclaimed to play a key role in observational learning of complexaction sequences by forming representations of the temporalordering of actions needed to guide subsequent performances(Frey and Gerry, 2006). We extend this finding by showing thatthis region is also engaged when the level of representationnecessary for retrieval of the correct answer is abstract.

What could explain the specific activation of early visual areasdepending on the way the visuomotor association was learned?Since conditions were identical in terms of complexity of thevisual stimuli, we suggest that the activation of the visual areasresults from top–down modulations. Indeed, results of severalneuroimaging studies have suggested that a frontal–parietalnetwork controls attention by sending “top–down” signalsmodulating the activity of the visual cortex during the executionof given tasks (Hopfinger et al., 2000; for review, see Corbetta andShulman, 2002). In the present study, attention to the hand is anintrinsic characteristic of learning by observation. The retrievalprocessmight necessitate the reactivation of the observedmove-ment during learning and the top–downmodulationmight affectdistinct visual areas depending onwhether the rule was learnedin LeO or in a TE context (Vidyasagar and Pigarev, 2007; Supèret al., 2001). In a recent study using magnetoencephalography,Nueuwenhuis et al. showed an increased gamma activity in thevisual areas (BA17 and BA18) when labile compared to stabilizedmemories are recalled (Nieuwenhuis et al., 2008). This resultfurther increases the evidence that the visual system is engagedin tasks beyond visual perception. These tasks include directedattention, working memory maintenance and long-term mem-ory encoding and recall (Jensen et al., 2007).

To conclude, we suggest that the mere perception of anabstract visual stimulus that has been previously associatedwith amotor response by trial and error or by observation of anactor is sufficient to trigger the activation of a set of brain areastypically involved in action observation or execution. Thissuggests the existence of a common neural system responsiblefor the storage of stimulus–response mappings learned eitherby trial and erroror byobservation of another's actions (Table 1).Retrieving an arbitrary visuomotor association learned byobservation engages the pars triangularis of the right prefrontalcortex, reflecting the potential value of other's actions in suchcontext and the need for its privileged processing.

The results of the present study try to bridge the gapbetween two widely explored cognitive functions: individuallearning and action observation and understanding. Visuo-motor individual learning is thought to be mediated by thefronto-striatal system (Wise andMurray, 2000; Hadj-Bouzianeand Boussaoud, 2003), whereas action observation engagesthe fronto-parietal mirror system. The fact that we can learnboth through trial and error and from observation of other'sbehavior suggests that these two systems interact to allow thetransfer of other's experience from the fronto-parietal systemto the fronto-striatal system. If so, the fronto-striatal and thefronto-parietal systems cooperate to ensure the learning ofbehavioral patterns via observation of other's actions.

Acknowledgments

Theauthorswish tothankB.Nazarian,M.Rothand J.L. Antonfor assistance with scanning, and Dr. G. Prabhu for correctingthe English. This work was funded by Action Concertée

Please cite this article as: Monfardini, E., et al., I learned from what yNeuroImage (2008), doi:10.1016/j.neuroimage.2008.05.043

Incitative NIC0050. EM is supported by funding from EC-contract number 27654 and AB is supported by a two-yearfellowship from the “Fondation pour la Recherche Médicale”.

Conflicts of interestThe authors have declared that no conflicts of interest exist.

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.neuroimage.2008.05.043.

References

Bandura, A., 1977. Self-efficacy: toward a unifying theory of behavioral change. Psychol.Rev. 84, 191–215.

Boyd, R., Richerson, P.J., 1985. Culture and the Evolutionary Process. University ofChicago Press, Chicago.

Boyd, R., Richerson, P.J., 1988. An evolutionary model of social learning: the effects ofspatial and temporal variation. In: Zentall, T.R., Galef, B.G. (Eds.), Social Learning: APsychological and Biological Approaches. Erlbaum, Hillsdale NJ, pp. 29–48.

Brovelli A., Laksiri N., Nazarian B., Meunier M., Boussaoud D., in press. Understandingthe Neural Computations of Arbitrary Visuomotor Learning Through fMRI andAssociative Learning Theory. Cereb. Cortex. PMID: 18033767.

Buccino, G., Binkofski, F., Fink, G.R., Fadiga, L., Fogassi, L., Gallese, V., Seitz, R.J., Zilles, K.,Rizzolatti, G., Freund, H.J., 2001. Action observation activates premotor and parietalareas in a somatotopic manner: an fMRI study. Eur. J. Neurosci. 13, 400–404.

Bunge, S.A., 2004. How we use rules to select actions: a review of evidence fromcognitive neuroscience. Cogn. Affect. Behav. Neurosci. 4, 564–579.

Bunge, S.A., Kahn, I., Wallis, J.D., Miller, E.K., Wagner, A.D., 2003. Neural circuitssubserving the retrieval and maintenance of abstract rules. J. Neurophysiol. 90,3419–3428.

Castro, L., Toro, M., 2004. The evolution of culture: from primate social learning tohuman culture. Proc. Natl. Acad. Sci. 101, 10235–10240.

Cavalli-Sforza, L.L., Feldman, M.W., 1981. Cultural Transmission and Evolution: AQuantitative Approach. Princeton Univ Press, Princeton.

Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-drivenattention in the brain. Nat. Rev. Neurosci. 3, 201–215.

Crone, E.A., Donohue, S.E., Honomichl, R.,Wendelken, C., Bunge, S.A., 2006. Brain regionsmediating flexible rule use during development. J. Neurosci. 26, 11239–11247.

Deiber, M.P.,Wise, S.P., Honda, M., Catalan, M.J., Grafman, J., Hallett, M., 1997. Frontal andparietal networks for conditional motor learning: a positron emission tomographystudy. J. Neurophysiol. 78, 977–991.

Deiber, M.P., Ibanez, V., Honda, M., Sadato, N., Raman, R., Hallett, M., 1998. Cerebralprocesses related to visuomotor imagery and generation of simple finger move-ments studied with positron emission tomography. NeuroImage 7, 73–85.

Donohue, S.E., Wendelken, C., Crone, E.A., Bunge, S.A., 2005. Retrieving rules forbehavior from long-term memory. Neuroimage 26, 1140–1149.

Eliassen, J.C., Souza, T., Sanes, J.N., 2003. Experience-dependent activation patterns inhuman brain during visual–motor associative learning. J. Neurosci. 23,10540–10547.

Frey, S.H., Gerry, V.E., 2006. Modulation of neural activity during observational learningof actions and their sequential orders. J. Neurosci. 26, 13194–13201.

Gazzola, V., Aziz-Zadeh, L., Keysers, C., 2006. Empathy and the somatotopic auditorymirror system in humans. Curr. Biol. 16, 1824–1829.

Gazzola, V., Rizzolatti, G., Wicker, B., Keysers, C., 2007. The anthropomorphic brain: themirror neuron system responds to human and robotic actions. NeuroImage 35,1674–1684.

Gallese, V., Goldman, A., 1998. Mirror neurons and the simulation theory of mind-reading. Trends Cogn. Sci. 2, 493–501.

Grèzes, J., Armony, J.L., Rowe, J., Passingham, R.E., 2003. Activations related to “mirror” and“canonical” neurons in the human brain: an fMRI study. NeuroImage 18, 928–937.

Hadj-Bouziane, F., Boussaoud, D., 2003. Neuronal activity in the monkey striatumduring conditional visuomotor learning. Exp. Brain Res. 153, 190–196.

Hopfinger, J.B., Buonocore, M.H., Mangun, G.R., 2000. The neural mechanisms of top–down attentional control. Nat. Neurosci. 3, 284–291.

Iacoboni, M., Molnar-Szakacs, I., Gallese, V., Buccino, G., Mazziotta, J.C., Rizzolatti, G.,2005. Grasping the intentions of others with one's ownmirror neuron system. PLoSBiol. 3, 529–535.

Jeannerod, M., 2001. Neural simulation of action: a unifying mechanism for motorcognition. NeuroImage 14, 103–109.

Jensen, O., Kaiser, J., Lachaux, J.P., 2007. Human gamma-frequency oscillationsassociated with attention and memory. Trends Neurosci. 30, 317–324.

Keysers, C., Wicker, B., Gazzola, V., Anton, J.L., Fogassi, L., Gallese, V., 2004. A touchingsight: SII/PV activation during the observation and experience of touch. Neuron 42,335–346.

Law, J.R., Flanery, M.A.,Wirth, S., Yanike,M., Smith, A.C., Frank, L.M., Suzuki,W.A., Brown,E.N., Stark, C.E., 2005. Functional magnetic resonance imaging activity during thegradual acquisition and expression of paired-associate memory. J. Neurosci. 25,5720–5729.

ou did: Retrieving visuomotor associations learned by observation,

7E. Monfardini et al. / NeuroImage xxx (2008) xxx–xxx

ARTICLE IN PRESS

Molnar-Szakacs, I., Iacoboni, M., Koski, L., Mazziotta, J.C., 2005. Functional segregationwithin pars opercularis of the inferior frontal gyrus: evidence from fMRI studies ofimitation and action observation. Cereb. Cortex 7, 986–994.

Molnar-Szakacs, I., Kaplan, J., Greenfield, P.M., Iacoboni, M., 2006. Observing complexaction sequences: the role of the fronto-parietal mirror neuron system. NeuroImage33, 923–935.

Nieuwenhuis, I.L., Takashima, A., Oostenveld, R., Fernández, G., Jensen, O., 2008. Visualareas become less engaged in associative recall following memory stabilization.Neuroimage 40, 1319–1327.

Rizzolatti, G., Craighero, L., 2004. The mirror-neuron system. Annu. Rev. Neurosci. 27,169–192.

Rizzolatti, G., Fadiga, L., Gallese, V., Fogassi, L., 1996. Premotor cortex and the recognitionof motor actions. Cogn. Brain Res. 3, 131–141.

Rizzolatti, G., Fogassi, L., Gallese, V., 2001. Neurophysiological mechanisms underlyingthe understanding and imitation of action. Nat. Rev. Neurosci. 2, 661–670.

Singer, T., Seymour, B., O'Doherty, J., Kaube, H., Dolan, R.J., Frith, C.D., 2004. Empathy forpain involves the affective but not sensory components of pain. Science 303,1157–1162.

Please cite this article as: Monfardini, E., et al., I learned from what yNeuroImage (2008), doi:10.1016/j.neuroimage.2008.05.043

Supèr, H., Spekreijse, H., Lamme, V.A., 2001. A neural correlate of working memory inthe monkey primary visual cortex. Science 293, 120–124.

Toni, I., Passingham, R.E., 1999. Prefrontal–Basal ganglia pathways are involved in thelearning of arbitrary visuomotor associations: a PET study. Exp. Brain. Res. 127, 19–32.

Toni, I., Ramnani, N., Josephs, O., Ashburner, J., Passingham, R.E., 2001. Learning arbitraryvisuomotor associations: temporal dynamic of brain activity. Neuroimage 14,1048–1057.

Vidyasagar, T.R., Pigarev, I.N., 2007. Modulation of neuronal responses in macaqueprimary visual cortex in a memory task. Eur. J. Neurosci. 25, 2547–2557.

Wicker, B., Perrett, D.I., Baron-Cohen, S., Decety, J., 2003a. Being the target of another'semotion: a PET study. Neuropsychologia 41, 139–146.

Wicker, B., Keysers, C., Plailly, J., Royet, J.P., Gallese, V., Rizzolatti, G., 2003b. Both of usdisgusted in My insula: the common neural basis of seeing and feeling disgust.Neuron 40, 655–664.

Wise, S.P., Murray, E.A., 2000. Arbitrary associations between antecedents and actions.Trends Neurosci. 23, 271–276.

Wise, S.P., di Pellegrino, G., Boussaoud, D., 1996. The premotor cortex and non-standardsensorimotor mapping. Can. J. Physiol. Pharmacol. 74, 469–482.

ou did: Retrieving visuomotor associations learned by observation,