Technology, expertise and social cognition in human evolution


Dr D. Stout, as above.


Paleolithic stone tools provide concrete evidence of major developments in human behavioural and cognitive evolution. Of particular interest are evolving cognitive mechanisms implied by the cultural transmission of increasingly complex prehistoric technologies, hypothetically including motor resonance, causal reasoning and mentalizing. To test the relevance of these mechanisms to specific Paleolithic technologies, we conducted a functional magnetic resonance imaging study of Naïve, Trained and Expert subjects observing two toolmaking methods of differing complexity and antiquity: the simple ‘Oldowan’ method documented by the earliest tools 2.5 million years ago; and the more complex ‘Acheulean’ method used to produce refined tools 0.5 million years ago. Subjects observed 20-s video clips of an expert demonstrator, followed by behavioural tasks designed to maintain attention. Results show that observational understanding of Acheulean toolmaking involves increased demands for the recognition of abstract technological intentions. Across subject groups, Acheulean compared with Oldowan toolmaking was associated with activation of left anterior intraparietal and inferior frontal sulci, indicating the relevance of resonance mechanisms. Between groups, Naïve subjects relied on bottom-up kinematic simulation in the premotor cortex to reconstruct unfamiliar intentions, and Experts employed a combination of familiarity-based sensorimotor matching in the posterior parietal cortex and top-down mentalizing involving the medial prefrontal cortex. While no specific differences between toolmaking technologies were found for Trained subjects, both produced frontal activation relative to Control, suggesting focused engagement with toolmaking stimuli. These findings support motor resonance hypotheses for the evolutionary origins of human social cognition and cumulative culture, directly linking these hypotheses with archaeologically observable behaviours in prehistory.


Neither toolmaking (Beck, 1980) nor cultural transmission (Whiten et al., 2007) is unique to humans. Yet there is a vast gulf between the accumulated (Tennie et al., 2009) complexity of human technology and that of any other living species. This disparity has been attributed to uniquely human physical (Johnson-Frey, 2003) or social (Tomasello et al., 2005) cognition, or both (Passingham, 2008).

Motor hypotheses of action understanding (Gallese & Goldman, 1998; Blakemore & Decety, 2001) suggest a possible unification of these explanations. The ‘Motor Cognition Hypothesis’ (Gallese et al., 2009) proposes that human social cognition has its phylogenetic and ontogenetic origins in ‘motor resonance’. Distinctive human capacities for technology, language and intersubjectivity might thus have a single origin in evolutionary modifications of a primate ‘mirror neuron system’ (Rizzolatti & Craighero, 2004). However, it is not clear to what extent action understanding relies on motor resonance as opposed to intention reading (Saxe, 2005; Grafton, 2009), nor at which level(s) action representations are shared between individuals (Jacob & Jeannerod, 2005; de Vignemont & Haggard, 2008).

To address these issues, we conducted a neuroimaging study in which human subjects observed the making of Paleolithic stone tools. Stone toolmaking is the earliest known uniquely human behaviour (Roux & Bril, 2005), dating back at least 2.6 million years (Semaw et al., 2003). Previous research (Stout & Chaminade, 2007) used FDG-positron emission tomography (PET) to study brain activation during stone toolmaking. In the earliest, ‘Oldowan’, technology a ‘hammerstone’ held in the dominant hand is used to strike sharp ‘flakes’ from a cobble ‘core’ manipulated by the other hand. We found this method to be associated with activation of parietal and frontal brain regions involved in sensorimotor coordination, grip selection and 3D shape perception. After that, 1.7 million years ago, more complex ‘Acheulean’ technology developed. Here cores were intentionally shaped into large cutting tools known as ‘handaxes’. We found this method to be associated with activation of the right inferior frontal gyrus (Stout et al., 2008), a region implicated in the hierarchical organization of action (Koechlin & Jubault, 2006).

In the present study we used functional magnetic resonance imaging (fMRI) to compare brain activation during the observation of Oldowan and Acheulean toolmaking. The Motor Cognition Hypothesis proposes that action understanding is tied to motor expertise (Gallese et al., 2009), but learning clearly requires understanding of actions not yet in the observer’s repertoire. Our design crossed observer expertise (Naïve, Trained, Expert) with technological sophistication (Oldowan, Acheulean) to examine the contribution of resonance and interpretation in understanding actions of varying familiarity and complexity. An account in terms of motor resonance predicts expertise effects in the putative human mirror neuron system (Rizzolatti & Craighero, 2004) and dorsolateral prefrontal cortex (Buccino et al., 2004; Vogt et al., 2007), regardless of complexity. An inferential account (Saxe, 2005) predicts complexity effects in brain regions associated with mental state attribution, including the medial prefrontal cortex (Frith & Frith, 2006). A mixed model (Grafton, 2009) makes less exclusive predictions, but might involve a shift from resonance to inference with increasing complexity and expertise.


Paleolithic toolmaking

The Paleolithic technologies investigated here are the same that were addressed in previous FDG-PET studies of subjects actually making stone tools (Stout & Chaminade, 2007; Stout et al., 2008). Oldowan flaking, known from approximately 2.6–1.6 million years ago, is a simple process of striking sharp cutting flakes from a stone core using direct percussion. However, even this simple technology requires substantial visuomotor coordination, including visual evaluation of core morphology (e.g. edge angles, location of convexities and concavities) in order to select appropriate targets for percussion, as well as active proprioceptive sensation and precise bimanual coordination to guide forceful blows to small targets on the core.

After approximately 1.7 million years ago, flake-based Oldowan technology began to be replaced by ‘Acheulean’ technology, involving the intentional shaping of cores into large cutting tools known as ‘picks’, ‘handaxes’ and ‘cleavers’. Such shaping requires greater perceptual-motor skill to precisely control stone fracture patterns and more complex action plans that relate individual flake removals to each other in pursuit of a distal goal. By 500 000 years ago, some Acheulean tools exhibit a high level of refinement that additionally requires the careful preparation of edges and surfaces, known as ‘platform preparation’, before flake removals. Platform preparation is often done on the face opposite a planned flake removal: the core is flipped over (‘inverted’) and a new hammerstone and/or hammerstone grip is selected and used to abrade/micro-flake the edge through light, tangential blows. This preparatory operation introduces a new sub-routine to toolmaking action plans, increasing their hierarchical depth. It is the ‘Late Acheulean’ method that is studied here.

As in previous FDG-PET studies, the current study also includes a control condition that consists of simple bimanual percussion of an unmodified core without any attempt to detach flakes. This condition is designed to include general demands of striking and manipulating a core, while omitting any more specific demands for percussive accuracy, core support, target selection and strategic planning involved in actual toolmaking.

Subjects and training

Three subject groups were included in the study, comprising technologically Naïve (n = 11), Trained (n = 10) and Expert (n = 5) individuals. All subjects were right-handed by self-report and had no history of neurological illness. The study was approved by the National Hospital for Neurology and Neurosurgery and the Institute of Neurology joint Ethics Committee. Twenty-one individuals with no prior experience of stone toolmaking were recruited via advertisements posted to electronic mailing lists maintained by the University College London Functional Imaging Lab and Institute of Archaeology. Respondents chose to participate in the Naïve or Trained group. Individuals who elected training attended 16 1-h training sessions over an 8-week period, in groups of 2–3 subjects per session. During training, subjects were provided with tools and raw materials for practice, as well as demonstrations and interactive verbal and gestural instruction by the first author. Subjects improved with training, but none achieved expertise in shaping handaxes (Supporting Information Fig. S1). Products of the 1st, 8th and 16th sessions of each subject were collected for further analysis (forthcoming). Because it became clear during training that subjects would not achieve the desired expertise, a group of five subjects with pre-existing expertise was included in the ongoing study. Expert subjects were drawn from the small extant community of academic and craft stone toolmakers, and were contacted directly. Imaging sessions for Naive, Trained and Expert subjects were interspersed over the course of the study.

Subjects in all groups received the same instructions before scanning, consisting of a scripted briefing, accompanying PowerPoint presentation, and Cogent script showing instructions and exemplar stimuli (not used in experiment) as presented in the scanner. Crucially, instructions included a description of the methods and aims of Paleolithic stone toolmaking so that even Naïve subjects had basic conceptual knowledge of the technology.


Twenty-second video clips (Supporting Information Video S1) were extracted from full-length videos of an expert toolmaker (right-handed) engaged in Oldowan flaking (n = 6), Acheulean shaping (n = 6) and the Control condition (n = 6). All videos were recorded on the same day with constant camera position and lighting. The demonstrator was seated facing the camera, and supported the core on his left thigh or above his lap in his left hand. The field of view included this workspace and the full range of arm movements, but did not extend to the face. Flint from a single quarry in Suffolk, UK was used for all toolmaking, and video segments were deliberately selected from early stages of flaking/shaping (e.g. prior to establishment of symmetrical ‘handaxe’ shape) so that size, shape, colour and other large-scale visual characteristics of cores did not differ systematically across stimulus types. Nevertheless, action sequences portrayed in the clips clearly reflected technological differences.

Nine types of technological action were identified in the videos, and their frequencies in the actual stimuli used recorded using the EthoLog 2.2.5 behavioural transcription tool (Table 1). These are: (i) percussive strikes with the right hand; (ii) shifts of the left-hand core grip; (iii) rotations of the core in the left hand; (iv) shifts of the right-hand hammerstone grip; (v) inversions (flipping over) of the core with the left hand; (vi) changing of the hammerstone (here the demonstrator reached off camera to exchange one hammerstone for another, see Supporting Information Video S1); (vii) abrasion/micro-flaking of core edges with right hand; (viii) sweeping of detached flakes and fragments off the thigh with the right hand (the hammerstone itself or an extended finger may be used); (ix) grasping of a detached flake or fragment with the right hand to remove it from the thigh, usually with a side-to-side ‘scissor’ grip of index and middle fingers, rarely (twice) with a pad-to-pad ‘pincer’ grip of thumb and index finger (Supporting Information Video S1). Importantly, the total number of actions declines from Control to Oldowan to Acheulean stimuli. This is also true for right- and left-hand actions considered separately. Thus, increasing activation across conditions must be explained in terms of the increasing diversity and causal/intentional complexity of actions rather than their simple quantity.

Table 1.   Frequency of technological actions in fMRI stimuli
Shift core gripL391213
Rotate coreL3716 7
Shift hammerstone gripR 3 0 8
Invert coreL 0 2 7
Change hammerstoneR 0 1 3
Abrade/micro-flakeR 0 012
SweepR 014 8
Grasp flakeR 0 8 3
Flakes detachedN/A 02916

There are four action types that are substantially more numerous in, or unique to, Acheulean stimuli: hammerstone grip shifts; hammerstone changes; core inversions; and abrasion/micro-flaking. These actions are all components of the distinctive ‘platform preparation’ operation discussed above, and their frequency directly reflects the greater technological complexity of Late Acheulean toolmaking. This complexity includes increased contingency on detailed variation in hammerstone properties, grips and gestures, and in core morphology, orientation and support, as well as a greater hierarchical depth of action planning.

fMRI paradigm

Subjects lay supine in the 3T Siemens Allegra MRI scanner at the Wellcome Trust Centre for Neuroimaging, pads positioned on the side of the head to reduce movement. Subjects underwent six sessions of approximately 7 min, and each session comprised 12 trials, corresponding to one repetition of six experimental conditions defined by a three × two factorial plan.

1. Stimulus: 20-s video clips of the Control stimulus, Oldowan or Acheulean toolmaking.

2. Task: following stimulus presentation, subjects were instructed either to simulate themselves continuing to perform the action they saw (Imagine) or to decide whether, in their opinion, the actor was successful in achieving his goal (Evaluate).

Prior to entering the scanner, subjects were instructed to watch each video ‘carefully’, to ‘try to understand what the demonstrator is doing’ and that after each video they would be ‘asked to do one of two things’, which were then explained. In the scanner, each trial was started by the presentation of the stimulus, followed by: (i) 1.5 s of a fixation cross; (ii) a written instruction indicating the Task (‘Imagine’ or ‘Evaluate’) that remained on screen for 5 s; and (iii) a response screen displaying the appropriate question (‘Did you finish?’ or ‘Was he successful?’). The side for yes and no responses was randomly assigned to the left and right button press and indicated by the position of the words ‘Yes’ and ‘No’ on screen. The response screen remained visible for 1.5 s or until subjects replied, and was followed by a fixation screen (minimum 1 s) for a total trial duration of 29 s. In addition, each session included four 12-s rest trials, each of which started with a 1-s ‘Rest’ indication, and ended with a 1-s ‘End of rest’ indication plus a 1-s fixation screen, giving a total duration of 15 s. Trials were interleaved so that in each session, experimental trials took place in blocks of two or three. Each session comprised one repetition of each of the six conditions; the order was pseudo-randomized to avoid repetition of the type of stimulus in consecutive trials and of individual stimulus videos within a session. Presentation of stimuli and recording of participants’ responses were carried out using Cogent ( running in Matlab 6.5 (MathWorks™).

fMRI recording and preprocessing

In each of the six experimental sessions, a T2*-weighted, gradient-echo, echo-planar imaging sequence was used to acquire 164 40-slice (2 mm thickness and 1 mm gap; TE = 65 ms; α = 90 °) volumes covering the whole brain and cerebellum with an in-plane resolution of 3 × 3 mm (64 × 64 matrix, fov 192 × 192 × 144 mm3; TR = 2600 ms). A high-resolution (1 × 1 × 1 mm3) structural image (MPRAGE sequence) was also collected.

fMRI data were analysed using SPM8 ( procedures, running in Matlab 7.6 (MathWorks™), after discarding the first four dummy volumes in each session to allow for T1 equilibrium effect. Slice timing correction was applied to correct for offsets of slice acquisition. EPI volumes were realigned to the first volume for each subject to correct for interscan movement, and unwarped for movement-induced inhomogeneities of the magnetic field using realignment parameters (Andersson et al., 2001). EPI volumes were stereotactically normalized into the standard space defined by the Montreal Neurological Institute (MNI) using a two-step procedure: the mean EPI image created during realignment was coregistered with the structural image, which was spatially normalized to the SPM T1 template using a 12-parameter affine and non-linear cosine basis function transformation, both transformations being subsequently applied to all EPI volumes. Normalized images were smoothed using an 8-mm isometric Gaussian kernel to account for residual inter-subject differences in functional anatomy (Friston et al., 2007).

fMRI statistical analysis

Analysis of the functional imaging data entailed the creation of statistical parametric maps representing a statistical assessment of hypothesized condition-specific effects (Friston et al., 1994). A random effect procedure was adopted for data analysis. Within individual subjects, the 20-s stimulations were modelled for the three types of stimuli (Control, Oldowan, Acheulean), the 5-s tasks were modelled for the three types of stimuli and two tasks (Imagine, Evaluate), and the motor responses were modelled as events (duration 0) irrespective of the experimental condition. Rest was modelled as a 12-s condition. Each condition was defined with a boxcar function convolved with SPM8 canonical haemodynamic response function to estimate condition-specific effects with the General Linear Model. Low-frequency drifts were removed by a high-pass filtering with a cut-off of 128 s.

The current analysis focused on the response to Acheulean, Oldowan and Control stimuli, with individual subjects’ statistical maps for the four contrasts [Toolmaking: (Acheulean–Control) + (Oldowan–Control); Acheulean–Control, Oldowan–Control, Acheulean–Oldowan] entered in second-level analyses of variance, with the group (Naïve n = 11, Trained n = 10, Expert n = 5) as between-subjects factor. As all groups comprise neurotypical adults, we hypothesized equal variance between populations in order to control for differences in group size (Penny & Holmes, 2003).

Common brain response irrespective of expertise was investigated using a minimum statistic conjunction (Nichols et al., 2005) between the three groups. Brain response specific of each group was assessed by masking exclusively the effect of this group by a global null conjunction (< 0.05 uncorrected) of the other two groups; for instance, the contrast between Acheulean and Oldowan in Naïve is exclusively masked by a conjunction of the same contrast in Trained and Expert subjects. Our procedure used exclusive masking instead of interactions, which were not significant at the threshold used, to favour the effects within the group of interest over the reversed effects in the other groups, which are included in the statistics of interactions (Culham, 2006). All contrasts were thresholded at P < 0.05 FDR-corrected with an extent threshold of 20 voxels.

Anatomical localization was performed using a brain atlas (Duvernoy, 1999) and, in particular for inferior frontal and parietal clusters, functional localization made use of distribution analysis of the activated voxels on the basis of probabilistic cytoarchitectonic maps (Eickhoff et al., 2007) implemented in SPM (Eickhoff et al., 2005). For the sake of consistency, only anatomical labels are used in the tables. Thus, clusters attributed to Brodmann area (BA) 44 were labelled ‘pars opercularis’ (Amunts et al., 1999), those attributed to BA45 were labelled ‘pars triangularis’ (Amunts et al., 1999), and those attributed to BA6 were labelled ‘precentral gyrus’ (Geyer, 2003). In the parietal cortex, clusters attributed to areas PF and PG (Caspers et al., 2006) were labelled ‘inferior parietal lobule’, and those attributed to hIP1 and hIP2 (Choi et al., 2006) were labelled ‘anterior intraparietal sulcus’. While recognizing that functional localization and anatomical landmarks may not strictly overlap in individuals, these conventions were adopted in the interest of coherence in the presentation of results. Statistical maps were rendered on FreeSurfer’s fsaverage pial surface with 50 inflation steps (

Local activity

In order to assess the effect of Group, local activity in clusters of interest was further characterized using the SPM extension toolbox MarsBar ( to extract percentage signal change in 5-mm radius volumes centred on the maximum of each cluster, then analysed with spss.



Common response

Across all subject groups, the contrast of Toolmaking conditions with Control yielded activations is a series of cortical regions, including a large cluster extending from the primary visual and lateral occipital cortices to the inferior temporal cortices, intraparietal sulci, inferior parietal cortices and postcentral gyrii bilaterally. In the frontal cortex, responses were found in right pars triangularis and bilateral pars opercularis of the inferior frontal gyrus, as well as in the dorsal precentral gyrii bilaterally (Fig. 1; Table 2).

Figure 1.

 Left: local brain activity in Toolmaking–Control irrespective of subject expertise (FDR P < 0.05, extent k > 20). Rendered with freesurfer on top, lateral and ventral views of the two hemispheres in neurological convention (details in Methods and Table 1). Right: percent signal change across the three technologies in the right pars triangularis (***< 0.001; *< 0.05).

Table 2.   Brain activity in response of the observation of Toolmaking compared with Control stimuli, common to the three groups (minimum statistic conjunction) and by subject expertise (exclusive masking)
Location xyzZ-scoreExtent
  1. All results are FDR P < 0.05, extent k > 20. All coordinates MNI.

Frontal lobe
 LeftDorsal precentral gyrus−32−8584.39421
 RightDorsal precentral gyrus30−8544.661003
 RightPars opercularis6010265.68747
 LeftPars opercularis−548224.89612
 RightPars triangularis5044103.2353
Parietal lobe
 LeftIntraparietal sulcus−26−52625.25
 LeftPostcentral gyrus−30−44625.9119 098
 RightIntraparietal sulcus32−60565.39
 RightPostcentral gyrus36−40565.06
 RightSupramarginal gyrus66−20365.82
 LeftSupramarginal gyrus−64−20365.46
Temporal lobe
 LeftInferior temporal gyrus−48−64−105.40
 LeftFusiform gyrus−48−50−205.57
 RightFusiform gyrus34−46−285.72
Occipital lobe
 RightMiddle occipital gyrus44−84125.23
 LeftCalcarine fissure−8−8443.40209
 RightMiddle occipital gyrus40−78−45.12
Basal ganglia
Naïve only
 LeftPrecentral gyrus−58−2303.8134
 RightPrecentral gyrus640283.8029
 LeftMiddle occipital gyrus−22−88144.7631
Trained only
 LeftPrecentral gyrus−18−10644.44242
 LeftMiddle frontal gyrus−244544.19
 LeftMedial frontal cortex−124483.95
 LeftPrecentral gyrus−38−6424.04
 RightIntraparietal sulcus44−48403.8331
 LeftPars opercularis−428304.19369
 RightPars opercularis6020123.78
 LeftPars triangularis−544084.81287
 RightAnterior insula342023.94165
 RightPars orbitalis4848−43.77107
 LeftMiddle temporal gyrus−60−62−64.0436
 RightInferior temporal gyrus62−56−104.2438
Experts only
 LeftSuperior parietal lobule−10−52744.46159
 RightPostcentral gyrus48−30623.8725

Two-way (Stimulus, Group) analysis of variance of the extracted percent signal change in the right pars triangularis revealed a main effect of Stimulus (< 0.001), with no effect of Group (= 0.9) or interaction between Stimulus and Group (= 0.5; see Fig. 1, right). All pairwise comparisons between stimuli are significant (Oldowan vs. Control P = 0.001; Acheulean vs. Control < 0.001; Acheulean vs. Oldowan P = 0.016).

Effect of expertise

The exclusive masking procedure used to isolate brain responses to the observation of Toolmaking stimuli unique to each level of expertise identified clusters (Fig. 2; Table 2) in the bilateral ventral precentral gyrus and left middle occipital gyrus in the Naïve group, and in the left superior parietal and right postcentral gyrus of Experts. Activations unique to the Trained group were much more numerous particularly in the frontal cortices, including medial frontal cortex, the right pars orbitalis, left pars triangularis, bilateral pars opercularis, right anterior insula, left posterior middle frontal gyrus and left precentral gyrus, as well as left middle temporal gyrus and right inferior temporal gyrus.

Figure 2.

 Local brain activity in Toolmaking–Control for Naïve (left), Trained (centre) and Expert (right) subjects (FDR P < 0.05, extent k > 20).

Acheulean vs. Oldowan

Common response

The minimum statistic conjunction between the three groups for the contrast Acheulean–Oldowan identified increases in activity in the anterior part of the left intraparietal sulcus (Fig. 3; Table 3), and in the left prefrontal cortex within the inferior frontal sulcus.

Figure 3.

 Left: local brain activity in Acheulean–Oldowan irrespective of subject expertise (FDR P < 0.05, extent k > 20). Right: percent signal change across the three technologies in the left anterior intraparietal sulcus (top) and inferior frontal sulcus (bottom; ***< 0.001; *< 0.05).

Table 3.   Brain activity in response of the observation of Acheulean compared with Oldowan toolmaking stimuli, common to the three groups (minimum statistic conjunction) and by subject expertise (exclusive masking)
  1. All results are FDR P < 0.05, extent k > 20. All coordinates MNI.

Location xyzZ-scoreExtent
 LeftAnterior intraparietal sulcus−50−36424.7453
 LeftPars triangularis−4632266.0294
Naïve only
 LeftSuperior frontal gyrus−2422564.2151
 LeftPars opercularis−6012244.4683
Experts only
 RightMedial frontal cortex642505.2524
 RightAnterior intraparietal sulcus46−48465.6544
 RightInferior parietal lobule62−52364.8022

In agreement with SPM whole-brain investigation, analysis of variance of activity extracted in these clusters indicated a main effect of the stimulus (both < 0.001), while there was no effect of Group or interaction between Group and Stimulus (all > 0.3) in these regions. Activity in Acheulean was significantly increased compared with Oldowan (< 0.001) and Control (< 0.05) for the left prefrontal cortex cluster, and all pairwise comparisons were significant (< 0.001) for the anterior intraparietal sulcus.

Effect of expertise

In Naïve subjects, there were activations for Acheulean–Oldowan in the left frontal cortex, dorsally in the superior frontal gyrus and ventrally in the pars opercularis of the inferior frontal gyrus (Fig. 4; Table 3). The latter activation was in a similar location to that previously reported for the actual performance (as opposed to observation) of stone toolmaking (Stout & Chaminade, 2007). No cluster survived the thresholds used in this analysis for Trained subjects.

Figure 4.

 Local brain activity in Acheulean–Oldowan for Naïve (left) and Expert (right) subjects (FDR P < 0.05, extent k > 20).

In Experts (Fig. 4; Table 3), there were clusters in the right medial frontal and parietal cortices. The latter were localized in the inferior parietal lobule, and in the anterior intraparietal sulcus area hIP1 (Choi et al., 2006; see also Jubault et al., 2007).


To identify brain systems involved in the observation of Paleolithic toolmaking, we examined contrasts of toolmaking observation with a control condition. Results were remarkably similar to those obtained from previous FDG-PET studies of toolmaking execution, despite the different experimental tasks and imaging modalities used. This indicates that neural systems involved in the observational understanding of Paleolithic toolmaking are very similar to those involved in execution.

To investigate the effects of expertise on toolmaking observation, we examined the unique responses of each subject group (Naïve, Trained and Expert) to Toolmaking stimuli. This provided evidence for the functional ‘reorganization’ (Kelly & Garavan, 2005) of activation between groups, reflecting expertise-dependent shifts in cognitive strategy.

To investigate the specific demands of understanding increasingly complex Paleolithic technologies, we examined the contrast in brain response to Acheulean vs. Oldowan stimuli in Naïve, Trained and Expert subjects. This revealed a significant main effect of technological complexity across groups, as well as distinct responses in the Naïve and Expert groups. The localization of these expertise-dependent effects suggests that stone toolmaking action understanding depends on a complex mixture of top-down, bottom-up, conceptual and embodied processes (cf. Grafton, 2009).

Contrasts of Toolmaking with Control condition

Common response

Contrasts of toolmaking stimuli with Control yielded activations in a series of cortical regions, notably including inferior frontal gyrus, dorsal premotor cortex, intraparietal sulcus and the inferior parietal lobule (Fig. 1; Table 1). Activations in these regions have commonly been reported in imaging studies of action observation (Grezes & Decety, 2001; Grafton, 2009; Caspers et al., 2010), and they are thought to comprise a network supporting action understanding through the covert simulation of observed behaviours. In keeping with this, the observed activations closely match (see also Supporting Information Fig. S2; Tables S1 and S2) those reported in previous FDG-PET studies, in which subjects actively produced tools rather than simply observing toolmaking (Stout & Chaminade, 2007; Stout et al., 2008).

Particularly notable is activation of the pars triangularis of the right inferior frontal gyrus. Pars triangularis activation is more typically associated with linguistic processing (e.g. Bookheimer, 2002; Musso et al., 2003), but has been reported during action observation (Johnson-Frey et al., 2003; Molnar-Szakacs et al., 2005; Caspers et al., 2010). It has been proposed (Rizzolatti & Craighero, 2004) that such activation reflects the ‘syntactic’ processing of hierarchically organized actions (cf. Koechlin & Jubault, 2006). This leads to the expectation that pars triangularis activity should respond to variation in the complexity of observed actions (Caspers et al., 2010). Such an effect of stimulus complexity is observed here (Fig. 1), in keeping with previous findings of pars triangularis activation during the execution of Acheulean, but not Oldowan, toolmaking (Stout et al., 2008; Table 2).

Effect of stone toolmaking method

Across groups, the increased technological complexity of Acheulean stimuli compared with Oldowan (Table 1) was associated with activation of the anterior intraparietal sulcus and inferior frontal sulcus, both in the left hemisphere (Fig. 3; Table 3). The anterior intraparietal sulcus is a core component of the putative human mirror neuron system (Grafton & Hamilton, 2007). It is thought to contribute to the understanding of ‘immediate’ action goals, such as grasping to eat vs. to place in macaque monkeys (Fogassi et al., 2005), or taking a cookie vs. a diskette in humans (Hamilton & Grafton, 2006).

In monkeys, the anterior bank of the intraparietal sulcus changes its connectivity and response patterns when the animals train to use tools (Hihara et al., 2006), enabling an integration of visual and somatosensory stimuli. This is argued to support tool use through assimilation of the tool into the monkey’s body schema (Maravita & Iriki, 2004), such that ‘tools become hands’ (Umiltàet al., 2008). However, human left anterior inferior parietal lobule displays a specific response to observed tool use (as opposed to unassisted manual prehension) that is absent in monkeys (Peeters et al., 2009). This suggests that hominoid anterior inferior parietal cortex may be evolutionarily derived to play a new role in coding the distinct functional properties of hand-held tools (Johnson-Frey et al., 2005; Peeters et al., 2009; Jacobs et al., 2010; Povinelli et al., 2010).

The centre of anterior inferior parietal cortex activation reported here is somewhat posterior (−50, −36, 42 vs. −52, −26, 34) to that of Peeters et al. (2009); however, extraction of the volume of interest used by Peeters et al. (coordinates from Orban, pers. comm.) confirms that the same effect of stimulus is indeed present in this region. This response to increasingly complex Paleolithic toolmaking is consistent with the hypothesis that human technological evolution was supported, at least in part, by the emergence of enhanced neural mechanisms for representing the causal properties of hand-held tools (Johnson-Frey, 2003; Wolpert, 2003; Peeters et al., 2009).

The main effect in the prefrontal cortex was centred on the inferior frontal sulcus. In macaques, this region is heavily interconnected with the anterior inferior parietal lobule (Pandya & Seltzer, 1982) and the parietal operculum (Preuss & Goldman-Rakic, 1989), in keeping with the co-activation observed here, and suggesting involvement in the integration of visuospatial and somatosensory information. In an fMRI study with macaques, there was activation in this area during the observation of actions (Nelissen et al., 2005). In contrast to more the posterior premotor cortex (F5c) where mirror neurons were originally recorded, the ventral prefrontal cortex also responded to abstract or context-free stimuli, including isolated hands, robotic hands and shapes (Nelissen et al., 2005), indicating representation and integration of actions at a relatively high level. In humans, activation of similar coordinates is reported during observation and preparation to imitate complex hand postures (guitar chords), perhaps indicating a role for this region in the selection and combination of motor elements into novel actions (Vogt et al., 2007).

It is thus likely that the increase in prefrontal activation for Acheulean–Oldowan reflects the greater temporal and relational complexity of Acheulean toolmaking actions, which, to a greater extent than Oldowan flaking, are organized into flexible and internally variable action chunks, such as ‘platform preparation’ vs. ‘primary flake removal’ (Pelegrin, 2005; Stout, 2011). No significant prefrontal activation was observed for Oldowan–Control, in keeping with previous conclusions regarding the relative simplicity of Oldowan action sequences (Stout & Chaminade, 2007; Stout et al., 2008).

On this interpretation, the anterior inferior parietal cortex and the inferior frontal sulcus form a parieto-frontal circuit involved in representing episode-specific intentions, causal relations and multi-component action sequences during toolmaking observation. The apparent abstraction (Hamilton & Grafton, 2006; Badre & D’Esposito, 2009) of causal/intentional processing in this circuit may be compared with a proposed ‘intermediate’ level representing ‘intentions in action’ as goal-oriented sequences of motor commands and predicted outcomes (de Vignemont & Haggard, 2008).

Expertise effects on response to stimuli

Varying expertise across subject groups was associated with qualitative shifts in the set of brain regions activated in response to Acheulean compared with Oldowan stimuli (Fig. 4; Table 3). These differences suggest a functional reorganization (Kelly & Garavan, 2005) involving the adoption of different cognitive strategies for action understanding. Naïve subjects show activation in core motor resonance structures together with the ventral prefrontal cortex, as expected for a low-level strategy of novel action understanding through kinematic simulation. Trained subjects show strong, statistically indistinguishable responses to both Oldowan and Acheulean stimuli, perhaps reflecting the particular social context and motivational set associated with training. Finally, Expert subjects display activation in the medial prefrontal cortex, a classic ‘mentalizing’ region, suggesting a relatively high-level, inferential strategy of intention reading.

Naïve subjects

One cluster exclusive to technologically Naïve subjects occurred in the pars opercularis of the left posterior inferior frontal gyrus (Fig. 4, left). Pars opercularis is another core component of the putative human mirror neuronal system (Rizzolatti & Craighero, 2004), which, in contrast with the performance-monitoring functions of the anterior inferior parietal cortex described above, is thought to be responsible for the generation of the kinematic models used to execute (Fagg & Arbib, 1998) or simulate (Carr et al., 2003; Grafton & Hamilton, 2007; Kilner et al., 2007) motor acts.

Also unique to Naïve subjects was activation of the superior frontal gyrus, anterior to the frontal eye field (Lobel et al., 2001). Activation of this area is associated with the selection among competing responses (Petrides, 2005), and the more superior portion activated here is especially involved in the spatial domain (Volle et al., 2008). During imitation, this region may serve to maintain a representation of the observed goal in short-term working memory for later execution (Chaminade et al., 2002). Co-activation of the superior frontal gyrus and posterior inferior frontal gyrus may thus reflect Naïve reliance on kinematic simulation and top-down direction of attention to task-relevant spatial cues. When combined with the anterior inferior parietal and ventral prefrontal activations observed across all groups, these Naïve activations match the general expectations of a simulation model of novel action understanding (Buccino et al., 2004; Vogt et al., 2007).

Trained subjects

No activations exclusive to Trained subjects were observed in the Acheulean–Oldowan contrast. Comparison with the numerous activations observed in the contrast of Toolmaking–Control for Trained subjects (Table 2; Fig. 2) indicates that this result derives from the presence of similar responses to Oldowan and Acheulean stimuli rather than from the absence of significant differences from Control. This is corroborated by the observation of similar activations in separate contrasts of Oldowan–Control and Acheulean–Control (Supporting Information Figs S3 and S4; Tables S1 and S2). The Trained response to both Oldowan and Acheulean stimuli includes: (i) clusters in the anterior insula, lateral premotor cortex, frontal eye field and supplementary eye field likely related to attentional and affective engagement with the stimuli; and (ii) ventral prefrontal clusters likely associated with parsing of observed action sequences.

Insular activations unique to Trained subjects are in an anterior region associated with interoception, subjective feeling and perceptual awareness (Kikyo et al., 2002; Ploran et al., 2007; Craig, 2009). Activations of the left medial frontal cortex (close to y = 0) and posterior middle frontal gyrus appear to fall within the supplementary and frontal eye fields (Tehovnik et al., 2000), functional regions associated with saccades, visual attention and visual learning (Tehovnik et al., 2000; Grosbras et al., 2005). Together with activation of the precentral gyrus, a region commonly recruited during action observation (Grezes & Decety, 2001; Caspers et al., 2010), these activations likely indicate intense engagement by Trained subjects with the Toolmaking stimuli. These effects of training were not predicted, but are consistent with the pragmatic social and motivational context created by the training programme.

Also unique to Trained subjects were inferior frontal gyrus activations of bilateral pars opercularis, left pars triangularis and right pars orbitalis. These are probably best understood in terms of the putative role of the inferior frontal gyrus in the multi-level processing of stimuli along a posterior to anterior gradient of increasing hierarchical complexity (Koechlin & Jubault, 2006; Caspers et al., 2010), and may reflect the intense processing of all Toolmaking stimuli by highly motivated Trained subjects.

Expert subjects

Activations exclusive to Expert subjects were observed in the medial frontal cortex, anterior intraparietal sulcus and inferior parietal lobule of the right hemisphere (Fig. 4, right). The medial frontal cortex is a core element in the network of brain regions associated with the attribution of mental states (Frith & Frith, 2006), suggesting that Expert subjects rely on top-down interpretation of the demonstrator’s intentions in order to differentiate Acheulean from Oldowan toolmaking. The activation is centred at the border between a posterior region associated with the attribution of ‘private’ action intentions and an anterior region associated with communicative intentions (Grèzes et al., 2004a,b; Amodio & Frith, 2006), in a position closely approximating that activated when mentalizing about the internal states of a dissimilar other (Mitchell et al., 2006). It may reflect inference about the private technological ‘prior intentions’ of the demonstrator (Chaminade et al., 2002), rather than meta-cognition about the demonstrator’s communicative intentions toward the observer (Amodio & Frith, 2006: 274).

Activation of the right anterior intraparietal sulcus in Experts is comparable to expertise effects found in studies of dance observation (Calvo-Merino et al., 2005, 2006; Cross et al., 2006). The more anterior location the current activation may reflect somatotopy of response to the observation of upper vs. lower limb actions (Buccino et al., 2001). This particular region of right anterior intraparietal sulcus has also been linked with the preparation of successive sensorimotor task-sets during action sequence execution (Jubault et al., 2007).

Also activated in Experts was a region of right inferior parietal lobule known to support the stimulus-driven allocation of spatial attention (Corbetta & Shulman, 2002; Mort et al., 2003) during visuospatial sequence learning (Rosenthal et al., 2009). This activation is posterior to the region associated with action outcome monitoring by Hamilton & Grafton (2008), and together with the right anterior intraparietal sulcus activation probably reflects Expert recognition of familiar toolmaking action sequences.

Broader implications

Contrasts with Control show that the observation of Paleolithic toolmaking recruits cognitive control mechanisms in the pars triangularis of the right inferior frontal gyrus, and that this response increases with the technological complexity of the observed actions. This matches results from earlier studies of subjects actively making stone tools, and is consistent with an evolutionary scenario in which manual and perceptual-motor adaptations were critical to the earliest stages of human technological evolution (Wynn & McGrew, 1989; Ambrose, 2001; Byrne, 2004; Bril & Roux, 2005; Stout & Chaminade, 2007), but later developments were more dependent on enhanced cognitive control (Faisal et al., 2010; Stout, 2010). These findings support long-standing intuitions regarding the cognitive sophistication of Acheulean technology (e.g. Oakley, 1954; Wynn, 1979; Gowlett, 1986), and specifically highlight the complex hierarchical organization (Holloway, 1969; Stout et al., 2008) of Acheulean action sequences. This interpretation is further supported by the main effect of stimulus in the anterior inferior parietal and ventral prefrontal cortices across subject groups.

Differing responses to stimulus complexity between groups provide insight into the effects of expertise on action observation strategies. Activations specific to Naïve subjects suggest a strategy reliant on kinematic simulation (inferior frontal gyrus) and the top-down direction of visuospatial attention (superior frontal gyrus). This supports an account of early observational learning in which simulation of low-level action elements interacts with representations of mid-level intentions in action to produce a ‘best-fit’ understanding of complex, unfamiliar actions (cf. Vogt et al., 2007).

Interestingly, Trained subjects responded equally to Oldowan and Acheulean stimuli, activating a set of frontal regions related to subjective awareness, visual attention and multi-level action parsing. This unexpected result may reflect a strong motivation to attend to, analyse and understand all Toolmaking stimuli, generated by the social and pragmatic context of being a ‘learner’ (cf. Lave & Wenger, 1991; Stout, 2002). There is increasing awareness of the importance of such social and affective dimensions in understanding human cognitive evolution (Holloway, 1967; Hare & Tomasello, 2005; Burkart et al., 2009; Stout, 2010).

Unlike Naive and Trained subjects, Experts recruited a mixture of bottom-up, familiarity-based posterior parietal mechanisms for visuospatial attention (right inferior parietal lobule) and sensorimotor matching (anterior intraparietal sulcus) with high-level inference regarding technological ‘prior intentions’ in the medial frontal cortex. In this context, shared pragmatic skills may provide the foundation for sharing of higher level intentions, in keeping with the Motor Cognition Hypothesis (Gallese et al., 2009). More broadly, the apparent shift in observation strategy from Naive kinematic simulation to Expert mentalizing is consistent with a ‘mixed’ model of action understanding (Grafton, 2009) involving contextually variable interactions between bottom-up resonance and top-down interpretation.

Complex, pragmatic skills like stone toolmaking can only be acquired through deliberate practice (Pelegrin, 1990; Whittaker, 1994) and experimentation (Ericsson et al., 1993), leading to the discovery of subtle causal relations that would remain ‘opaque’ (Gergely & Csibra, 2006) to observation and simulation alone. Mid-level ‘intentions in action’ represented in the anterior inferior parietal and the ventral prefrontal cortices, though likely to be inaccurate at first, appear to be important across skill levels and may play an important role in guiding such practice, perhaps contributing to the high fidelity of human social learning (the ‘ratchet effect’: Tomasello, 1999; Tennie et al., 2009). The effect of Toolmaking complexity in the anterior inferior parietal lobule in particular suggests that this phylogenetically derived (Peeters et al., 2009) region may have played a key role in human technological evolution 2.6–0.5 million years ago.


This research was funded by European Union project HANDTOMOUTH. We thank Bruce Bradley for acting as the expert demonstrator, and Stefan Vogt and an anonymous reviewer for helpful comments.


Brodmann area


functional magnetic resonance imaging


positron emission tomography