Action execution–perception links (mirror mechanism) have been repeatedly suggested to play crucial roles in social cognition. Remarkably, the designs of most studies exploring this topic so far excluded even the simplest traces of social interaction, such as a movement of the observer toward another individual. This study introduces a new design by investigating the effects of camera movements, possibly simulating the observer's own approaching movement toward the scene. We conducted a combined high-density EEG and behavioral study investigating motor cortex activation during action observation measured by event-related desynchronization and resynchronization (ERD/ERS) of the mu rhythm. Stimuli were videos showing a goal-related hand action filmed while using the camera in four different ways: filming from a fixed position, zooming in on the scene, approaching the scene by means of a dolly, and approaching the scene by means of a steadycam. Results demonstrated a consistently stronger ERD of the mu rhythm for videos that were filmed while approaching the scene with a steadycam. Furthermore, videos in which the zoom was applied reliably demonstrated a stronger rebound. A rating task showed that videos in which the camera approached the scene were felt as more involving and the steadycam was most able to produce a visual experience close to the one of a human approaching the scene. These results suggest that filming technique predicts time course specifics of ERD/ERS during action observation with only videos simulating the natural vision of a walking human observer eliciting a stronger ERD than videos filmed from a fixed position. This demonstrates the utility of ecologically designed studies for exploring social cognition.
Previous research repeatedly showed that a number of human cortical areas, including the ventral part of the precentral gyrus, the posterior part of the inferior frontal gyrus, and the rostral part of the inferior parietal lobule, are activated during both the execution and the observation of goal-related actions (Frith & Frith, 2010; Rizzolatti & Sinigaglia, 2010; Cattaneo & Rizzolatti, 2009; Hari & Kujala, 2009; Decety & Grèzes, 2006; Gallese, Keysers, & Rizzolatti, 2004; Rizzolatti & Craighero, 2004). Furthermore, even the observation of the outcomes of certain hand-related actions was found to activate observers' motor cortex (see Heimann, Umiltà, & Gallese, 2013; Longcamp, Anton, Roth, & Velay, 2005, regarding written language symbols and Sbriscia-Fioretti, Berchio, Freedberg, Gallese, & Umiltà, 2013; Umiltà, Berchio, Sestito, Freedberg, & Gallese, 2012, regarding abstract artworks). These findings were interpreted as evidence for the existence of an action execution–perception link crucially entangling action and perception, generally defined as the mirror mechanism.
Interestingly, studies investigating the precise circumstances of the visuomotor activations described above indicated that they are sensitive to, as being modulated by, the observer's motor experience with the observed action as well as by his or her body's actual spatial position with respect to the observed action. First, it was reported that the activation of motor areas during action perception is significantly higher when the observed actions belong to observers' motor repertoire. Calvo-Merino, Glaser, Grèzes, Passingham, and Haggard (2005) and Orgs, Dombrowski, Heil, and Jansen-Osman (2008) showed this effect in expert dancers observing dancing steps of their own professional style in comparison with dancing steps with which they did not have a professional experience (Classic Ballet vs. Capoeira). Aglioti, Cesari, Romani, & Urgesi (2008) found similar results comparing elite basketball players and nonathletes during observation of basket shots. These results were interpreted as indicating that motor activation found during action perception is modulated by motor familiarity with these actions (see, e.g., Calvo-Merino, Grèzes, Glaser, Passingham, & Haggard, 2006). Second, experiments in monkeys showed that the discharge of premotor mirror neurons is modulated by the distance between the observing monkey and the observed agent. Some neurons showed stronger activation when the observed action occurred within monkey's peripersonal space, whereas others responded only to observed actions performed within monkey's extrapersonal space (Caggiano, Fogassi, Rizzolatti, Thier, & Casile, 2009).
It has been repeatedly suggested that action execution–perception links make up a crucial basis of social cognition. A core proposal in this context was that motor activation elicited by action-related visual stimuli might serve an “embodied simulative function” crucial for the understanding of the behavior of others, one of the main conditions for successful social interaction (see Gallese & Sinigaglia, 2011). In front of this background, the reported context sensitivity of the visuomotor activations is likely a result of the needs of specific social interactions.
Nevertheless, it is remarkable that the designs of studies used to investigate the underlying mechanisms of action execution–perception links are so often constrained or artificial in regards to their social/interactional character. In fact, they mostly avoid any traces of real social interactions, including any movement of the observer toward or away from the observed agent. Typically, participants are asked to move as little as possible while observing short video clips on a computer screen placed at a fixed distance in front of them to avoid artifacts and guarantee reproducibility. We considered it necessary to take possible steps toward more ecologically valid study designs. With this in mind, we used high-density EEG to determine whether various types of camera movements, more or less simulating an observer's own movement toward the observed acting agent, might modulate observer's mirror mechanism.
Specifically, we studied the modulation of the mu rhythm as a common marker of the mirror mechanism in humans. Previous studies showed that voluntary action execution and observation correlate with event-related desynchronization (ERD) in upper alpha bands as well as in lower beta bands recorded over sensorimotor areas (Perry, Troje, & Bentin, 2010; Pfurtscheller & Lopes da Silva, 1999; Leocani, Toro, Manganotti, Zhuang, & Hallet, 1997; Stancak & Pfurtscheller, 1996; Toro et al., 1994; Derambure et al., 1993; Pfurtscheller & Berghold, 1989; Pfurtscheller & Aranibar, 1979). It was also shown that this ERD can be modulated by contextual factors such as familiarity of the observed action (Orgs et al., 2008).
Building on the design normally employed to investigate the hand action mirror mechanism, we focused on two questions: (1) Whether the mirror mechanism responds differently to the observation of the same hand action filmed by a static camera in comparison with a moving camera approaching the scene. (2) Whether the activation in focus is modulated by different ways a camera can be used to approach the scene. More precisely, is the mirror mechanism differently modulated by camera movements such as (a) zooming in on the scene, (b) real camera movement toward the scene realized by using a dolly (camera mounted on fixed tracks), and (c) real camera movement toward the scene obtained by using a steadycam (camera fixed to the body of the cameraman, walking toward the scene)?
We also investigated whether differences among viewing conditions (still, zoom, dolly, steadycam) could be related to participants' subjective reports regarding the feeling of involvement and the experienced naturalness or artificiality of the camera movement used.
Nineteen healthy volunteers, recruited by public announcement, participated in the experiment. Two participants were subsequently excluded from analysis because of artifacts in the EEG data/lack of typical ERD pattern (see EEG Recording and Analysis section). Among the remaining participants, 7 were men and 10 were women, mean age was 22.8, and all were right-handed as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971). All participants were paid 25 Euros for reimbursement. Before the experiment, they received written and oral experimental instructions. After the experiment, each participant was debriefed. Written informed consent was obtained from all participants before entering the study. The study was approved by the local ethical committee.
Stimuli consisted of short video clips of 3-sec length showing an agent (one woman, one man) grasping an object from a table placed in front of the agent. The background consisted of a black wall in front of which we placed a panel with a black and white geometrical pattern to enforce the 3-D perception of the room. Grasped objects included a marble, a battery, an eraser, a scotch tape, an espresso cup, a plastic mug, a plastic ball of 8-cm radius, and a packet of tissues (see Figure 1).
Video clips were recorded in a professional film studio, enabling us to film the same scene four times under highly controlled conditions. The camera starting position was always 260 cm from the object; the end position (in case of movement) was 80 cm from the object. The camera movement speed and its height from the ground were kept identical in the three different movement conditions, so that the only difference among them consisted of the type of movement used to approach the scene: zoom, dolly, and steadycam. Figure 2 shows four still frames taken from a video clip filmed with a steadycam, including start and end positions.
The experiment consisted of two different parts comprising (1) a 50-min EEG recording session and (2) a 10-min rating task.
The EEG was recorded during five blocks of about 10 min length each. After each block participants were given the possibility of a short break. Each block consisted of 80 trials (16 per condition—still, zoom, dolly, steadycam, and action execution, for details, see below). Participants were seated in an isolated EEG lab in front of a computer screen placed on a table at a distance of 50 cm. Participants were then instructed about the experimental procedure: Each trial began with a fixation cross of 200 msec, followed by a video (of 3-sec length each, presented in random order). In 80% of the trials, after stimulus presentation, a gray screen was displayed for 5 sec (guaranteeing the return of brain activity to baseline). Participants were asked to blink only in the second half of the gray screen period to minimize movement artifacts also in the resynchronization phase. In the remaining 20% of the trials (action execution condition), after stimulus presentation and before the gray screen, in addition the photo of an object (see above) appeared, and participants were asked to tell whether the object was the same they just had seen being grasped in the video displayed before. The answer had to be given by clicking the mouse with their right index finger. The mouse was positioned on the table in front of participants at a distance of 15 cm from their right hand that was kept at a fixed position indicated by a physical marker. When the object photo appeared, participants were asked to move their hand from start position to the mouse and click either the left button of the mouse, when seeing the same object as in the video just displayed before, or the right button of the mouse, when seeing a different object. They were furthermore asked to move back the hand to start position right after the click. If participants gave a wrong answer or did not answer within 3 sec, they were told the trial was incorrect/the answer given too slow, and the trial got repeated. This action execution condition served both as control for attention and to record participants' ERD during action execution (for experimental paradigm, see Figure 3).
The second part of the experiment consisted of a rating task. For this task, of the previous 64 video clips, 12 were chosen again (three per condition). Participants, still sitting in front of the screen as during the EEG recording session, were again shown these video clips and for each were asked six different questions (in six separate blocks, one for each question, always conducted in the same order):
1. How much did you feel involved in the scene?
2. How much did you feel like the actor?
3. How much did you feel as if you yourself would approach the scene? (was not asked for still camera video clips)
4. How comfortable did you feel watching the scene?
5. How realistic did you find the camera movement? (was not asked for still camera video clips)
6. How much did you feel the camera movement resembled a person's movement when approaching the scene? (was not asked for still camera video clips)
The first three questions were designed to investigate participants' potential feeling of involvement with the observed scene either in terms of empathy with the actor or in terms of embodiment with the camera. The fourth question was designed to explore how at ease participants were with the different ways of filming the scene. The last two questions were designed to measure participants' estimation of the ecological plausibility of the different types of camera movements with respect to those of a real observer approaching the scene. The rating was given by using the mouse to place a cursor on a 0–100 rating scale below each stimulus.
EEG Recording and Analysis
EEG data were acquired by a 128-channel Sensor Net (Electrical Geodesic, Eugene, OR) and recorded within standard EGI package Net Station 4.3.1. EEG was sampled at 250 Hz and band-pass filtered at 0.3–100 Hz, and electrode impedance was kept less than 50 kΩ (controlled after each block). The raw EEG data were recorded with the vertex (Cz) as the online reference and rereferenced offline to the common average (Muthukumaraswamy, Johnson, & McNair, 2004). Stimuli were presented with E-Prime 2.0 (Psychology Tools, Inc., Pittsburgh, PA), and at the beginning of each trial, all event markers were sent to Net Station. Participants' motion was monitored by the experimenter and video-recorded for offline analysis; if participants moved during the observation or rest conditions, the trial was excluded from further data analysis. EEG data were filtered offline with band-pass filter 0.3–30 Hz and segmented into specific time epochs. From observation trials, the whole 3 sec of stimulus presentation plus the first 2 sec of gray screen (resynchronization phase, see below) were analyzed. As baseline were taken 1000 msec of gray screen ending 1 sec before the start of the new trial (appearance of the fixation cross) in the observation trials. From action execution trials, segments of 1000 msec were cut, starting 500 msec before the motor response (button press) and ending 500 msec after it. Only the trials in which participants responded correctly were analyzed. The trials in which participants produced eye blinks and movement artifacts were rejected on the basis of the artifact detection tool supplied by Net Station and on the basis of a subsequent careful visual inspection of each segment. A minimum number of 50 trials for each condition was kept (fulfilled by all but one participant, who was consequently excluded from further analysis).
The time–frequency analysis was performed by continuous Morlet wavelet transformation in 0.5-Hz intervals in the frequency range from 1 to 30 Hz. Frequency–power coefficients were calculated by taking the average across trials. The wavelet transformation was calculated separately for each participant in all 128 channels for each condition. It was scaled by division using 250 msec out of the preceding fixation cross period.
Statistical analysis was performed on a selected cluster of eight electrodes in each hemisphere located around standard C3 and C4 sites (Electrodes 30, 31, 36, 37, 41, 42, 53, 54 left and 79, 80, 86, 87, 93, 103, 104, 105 right, as used in prior studies, see Streltsova, Berchio, Gallese, & Umiltà, 2010; Muthukumaraswamy & Johnson, 2004a, 2004b; Muthukumaraswamy et al., 2004; Bernier, Dawson, Webb, & Murias, 2007).
Data were analyzed regarding frequency power changes of the different components of the rolandic mu rhythm, reported to be modulated by voluntary action execution as well as action observation (Perry et al., 2010; Pfurtscheller & Lopes da Silva, 1999; Leocani et al., 1997; Stancak & Pfurtscheller, 1996; Toro et al., 1994; Derambure et al., 1993; Pfurtscheller & Berghold, 1989; Pfurtscheller & Aranibar, 1979). As it was shown that also, the event-related resynchronization (ERS) following ERD of the rolandic mu rhythm shows a specific pattern that can be modulated by contextual conditions (Heimann et al., 2013; Pfurtscheller, Neuper, Brunner, & Lopes da Silva, 2005; Muthukumaraswamy et al., 2004), we analyzed four different consequent time windows representing early ERD, beginning of ERS, rebound (oversynchronization), and return-to-baseline stages (see Statistical Analysis section for details).
For each participant specific alpha frequency bands were selected in the range of 8–14 Hz following the procedure described in previous studies (Babiloni et al., 2009; Oberman, McCleery, Ramachandran, & Pineda, 2007; Oberman, Pineda, & Ramachandran, 2007). The individual peak (F) of attenuated frequency was determined by calculating the ratio between the frequency power in action execution trials and during baseline in the six following subfrequency bands: 8–9, 9–10, 10–11, 11–12, 12–13, 13–14 Hz. Each value was then transformed into a log ratio, and the frequency that corresponded to the log ratio with the most negative value was taken as F. A 3-Hz range frequency band was chosen for each participant (F − 1; F + 1). For the following statistical analyses, the frequency power in this 3-Hz range was extracted in all conditions (number of participants selected per range: 7–10 Hz: 0; 8–11 Hz: 6; 9–12 Hz: 2; 10–13 Hz: 8; 11–14 Hz: 1; 12–15 Hz: 0). The data of a second participant for which no significant difference between baseline and action execution condition was found in any of the checked frequency bands were excluded from further analysis.
Because the central alpha frequency band (8–14 Hz) overlaps with the posterior alpha band, it is possible that recordings in central areas might be affected by this posterior activity. To check whether the alpha recorded in central areas was affected by posterior alpha, for the alpha range selected for each participant, we performed an additional analysis in four electrodes per hemisphere in occipital areas (electrodes 69, 70, 73, 74 in left occipital lobe and electrodes 82, 83, 88, 89 in right occipital lobe) using the same frequency bands as previously described.
Furthermore, in every participant three beta frequency ranges were analyzed (a lower band of 14–20 Hz, a middle band of 18–24 Hz, and an upper band of 24–30 Hz) using the same central electrode cluster as for the alpha range (regarding range selection, see Avanzini et al., 2012).
On the results of the statistical extraction, we applied an outlier detection (±2 SDs from mean) revealing outlier values in zoom condition of one participant. Such outlier values were replaced with the corresponding average values of the remaining participants.
To assess central alpha and beta desynchronization in sensory motor areas during different observation and action execution trials, we compared the frequency power extracted from wavelet for the different conditions using several ANOVAs.
1. To generally assess central alpha and beta ERD in sensory motor areas, for every band range considered (alpha, low beta, middle beta, high beta), we compared the frequency power extracted from wavelet during baseline (average of 1000 msec of period of gray screen ending 1000 msec before fixation cross) with its value during observation conditions (average of 3000 msec of video presentation) and action execution condition (average of 1000 msec: 500 msec before and 500 msec after button press) using a repeated-measures 2 × 6 ANOVA with two levels of Hemisphere (right vs. left) and six levels of Condition (baseline, four observation conditions [still, zoom, dolly and steadycam], and action execution).
2. To assess the time course of central alpha and beta ERD/ERS in sensory motor areas during the four observation conditions, for every band range considered (alpha and low beta—middle beta and high beta were left out due to results of Analysis I, see below) we first analyzed frequency power in 20 separate epochs of 250-msec length each. Because of the typical ERD/ERS/rebound pattern seen in the descriptive analysis of this analysis, we then chose four time windows to assess statistical differences among conditions in the different stages of the event-related modulation of the mu rhythm. The selected four windows were ERD window, consisting of the first second of stimulus presentation; ERS window, consisting of the second and third second of stimulus presentation; Rebound window, consisting of the first second after stimulus offset; Return to baseline window, consisting of the second second after stimulus offset. With the values of the power analysis for these four windows, we performed a repeated-measures 2 × 4 × 4 ANOVA with two levels of Hemisphere (right vs. left), four levels of Condition (still, zoom, dolly, steadycam), and four levels of Time (see above). To keep the relation to baseline in the picture, values used for this ANOVA were the log values of the condition/baseline division.
3. To control for effects in occipital electrodes for the alpha band, both ANOVAs described above were repeated for the occipital electrodes in the same frequency ranges.
Results were analyzed using a repeated-measures ANOVA for each question with the single main factor of Condition.
In all performed ANOVAs (of EEG analysis and rating task), we applied Duncan post hoc tests to further explore significant factors and interactions. Reported results are automatically adjusted for multiple comparisons. Error bars in all the graphs represent standard errors. In Figures 6 and 7, conditions are represented next to each other for the ease of visualization.
To generally assess central alpha ERD in sensory motor areas, we compared the frequency power for selected alpha frequency ranges extracted from wavelet during baseline with observation conditions (still, zoom, dolly, steadycam) and action execution condition. Descriptive analysis (see Figure 4) showed that, compared with baseline, ERD was present in all four observation conditions as well as during action execution (button press), with a maximum in the latter condition. A 2 × 6 ANOVA (Hemisphere × Condition) showed only a significant main effect for Condition, F(5, 80) = 17.882, p < .001.
Post hoc comparisons showed that frequency power for baseline was significantly higher than all other conditions and that frequency power for action execution was significantly lower than all other conditions (for all these comparisons p < .001). Differences between different observation conditions were not significant (for all ps > .3).
To control for similar effects in occipital regions, we repeated the analysis just described in occipital electrodes. The 2 × 6 ANOVA with two factors of Hemisphere and six factors of Condition (baseline, still, zoom, dolly, steadycam, and action execution) in the occipital region showed a main effect of Condition, F(5, 80) = 21.08, p < .001. Post hoc comparisons showed that frequency power for baseline was significantly higher than all other conditions (for all ps < .001), whereas there was no significant difference among observation conditions and action execution condition. These results support the notion that alpha ERD in occipital regions is discriminable from alpha ERD in central regions.
The ANOVA also showed a significant effect of Hemisphere × Condition, F(5, 80) = 3.3, p < .01. Post hoc comparisons showed that this difference was because of differences of the baseline values only, with the right hemisphere having a higher baseline value than the left hemisphere (p < .01). We therefore computed the lateralization index (Baseline right − Baseline left / (Mean of Baseline right and left)) for all participants in central and occipital electrodes for this condition and applied a paired sample t test to check for differences between the two regions. Descriptive analysis showed that the mean lateralization index of baseline values in central electrodes was 0.082, whereas the mean lateralization index of Baseline values in occipital electrodes was 0.096 (indicating a slightly higher lateralization to the right hemisphere in Occipitals). However, the paired sample t test did not show a significant difference between the two regions (t(16) = −0.240921, p > .8). This means that our data cannot support a hemispheric difference between the two ROIs in the occipital cortex regarding the mean power frequency over the whole time of video observation.
To generally assess beta ERD in central sensory motor areas (as part of the mu rhythm), we compared the frequency power of three different beta ranges (low, middle, and high beta) extracted from wavelet during baseline with that during observation conditions and that during action execution (see Figure 5).
Descriptive analysis for the low beta range (14–20 Hz; see Figure 5A) showed that compared with baseline ERD was present in all four observation conditions as well as during action execution, with a maximum in the steadycam condition. A 2 × 6 ANOVA (Hemisphere × Condition) accordingly showed a significant main effect for Condition. Post hoc comparisons revealed that the significant differences lay between baseline and all other conditions (p < .001), as well as between steadycam and action execution (p < .01).
Descriptive analysis for the middle beta range (18–24 Hz; see Figure 5B) as well as for the high beta frequency range (24–30 Hz; see Figure 5C) showed no ERD for observation or action execution, and the 2 × 6 ANOVAs (Hemisphere × Condition) showed no significant main effect for Condition or interaction.
Because of these results, we decided to further investigate only two ranges that gave significant results: the selected alpha frequency ranges and the low beta frequency range (14–20 Hz).
To assess the precise time course of ERD/ERS in sensory motor areas during the four different observation conditions in the frequency ranges of interest, we compared the log values of the frequency power of the chosen frequency ranges extracted from wavelet during the different observation conditions divided by baseline. For each frequency range, a 2 × 4 × 4 ANOVA design was created with 2 levels of Hemisphere (left vs. right), 4 levels of Condition (still, zoom, dolly, steadycam) and 4 levels of Time (4 time windows representing early ERD, beginning of ERS, rebound [oversynchronization], and return-to-baseline stages). For a better illustration of the time course, in the graphs we show the 20 epochs separately, marking the selected windows on the bottom.
The results of the 2 × 4 × 4 ANOVA for the selected alpha frequency ranges revealed a significant main effect of Time, F(3, 48) = 20.1, p < .001, as well as significant interactions of Condition × Time, F(9, 144) = 1.97, p < .05. Descriptive analysis regarding the main effect of Time showed a typical ERD-ERS-rebound pattern (see also Avanzini et al., 2012).
Descriptive and post hoc analysis regarding the interaction of Condition × Time showed the following characteristics (see Figure 6). (1) In the ERD phase (first second of video observation) in the selected alpha range, no significant differences among conditions were observed. (2) In the following time window of further ERD/beginning ERS (second and third second of video observation) conditions differentiated. Descriptive analysis showed that ERD was strongest for the steadycam, followed by dolly, then zoom, then still condition. Post hoc comparisons showed that differences were significant between still and steadycam condition (p < .001), zoom and steadycam condition (p < .05), and between still and dolly condition (p < .05). (3) In the rebound window (first second after stimulus offset) order of power values changed with descriptive analysis showing that zoom had the highest power, followed by dolly, then still, then steadycam. Post hoc comparisons showed that differences were significant between zoom and still (p < .01) and zoom and steadycam (p < .05). (4) In the returned-to-baseline window (second second after stimulus offset) no significant differences between conditions were measured.
To control for effects in occipital regions, we repeated the analysis done in central electrodes for occipital electrodes. The 2 × 4 × 4 ANOVA (2 levels of Hemisphere (left vs. right), 4 levels of Condition (still, zoom, dolly, steadycam), and 4 levels of Time (see above) showed a main effect of Hemisphere, F(1, 16) = 4.74, p < .05, a main effect of Time, F(3, 48) = 66.17, p < .001, and a significant Hemisphere × Time interaction, F(3, 48) = 4.98, p < .01. No effect was found regarding Condition × Time (p > .4).
Descriptive analysis regarding the effect of hemisphere showed that, as already reported for the 2 × 6 ANOVA, ERD was stronger in the right hemisphere. Descriptive analysis regarding the effect of Time showed an ERD pattern as expected (because of the presentation of a visual stimulus). Descriptive analysis regarding the effect of Hemisphere × Time showed that ERD was stronger in the right hemisphere for the first three time windows. Post hoc comparisons showed that all of these differences were significant (for Windows 1 and 2 p < .001, for Window 3 p < .05).
Because of these results, we repeated the comparison of the lateralization index for occipital and central electrodes. For this, for each participant and each region, we took the mean value of alpha frequency power over the four conditions of the right hemisphere in each time window, substracted the mean value of alpha frequency power over the four conditions of the left hemisphere, and divided the result by the mean value of these two numbers. Then, with the results of this calculation, we conducted a repeated-measure 2 × 4 ANOVA with 2 factors of Region and 4 factors of Time.
Still, as results did not show any significant Condition effect or Condition × Time interaction, they support the notion that alpha ERD in occipital cortices is discriminable from alpha ERD in motor areas, showing only effects likely because of visual attention without differing results for the conditions of interest.
Low Beta Range (14–20 Hz)
The results of the 2 × 4 × 4 ANOVA for the low beta range (14–20 Hz) revealed a significant main effect of Time, F(3, 48) = 41.03, p < .001, as well as a significant interaction Condition × Time, F(9, 144) = 2.02, p < .05. Descriptive analysis regarding the main effect of Time showed a typical ERD-ERS-rebound pattern (see also Avanzini et al., 2012).
Descriptive and post hoc analysis regarding the interaction of Condition × Time showed the following characteristics (see Figure 7). (1) In the beta range, already in the first ERD phase (first 1000 msec of stimulus presentation) conditions differed. Descriptive analysis showed that early ERD was strongest for steadycam, followed by dolly, then zoom, then still condition. Post hoc comparisons showed that differences were significant between still and steadycam (p < .05), zoom and steadycam (p < .05), and dolly and steadycam (p < .05). (2) Also in the following time window of further ERD/early ERS conditions differentiated. Descriptive analysis showed that ERD was still strongest for the steadycam, followed by dolly, then zoom, then still condition. Post hoc comparisons showed that differences were significant between still and steadycam condition (p < .001), still and dolly condition (p < .05), zoom and steadycam condition (p < .05), as well as dolly and steadycam condition (p < .01). (3) In the rebound window (first second after stimulus offset) the order changed. Descriptive analysis showed that zoom had the highest power, followed by dolly then still, then steadycam. Post hoc comparisons showed that differences were significant between zoom and steadycam (p < .05) and dolly and steadycam (p < .05) condition. (4) In the returned-to-baseline window no significant differences among conditions were measured.
Results of the rating task for Question 1 (see Figure 8A) showed that participants felt more involved in the scene when the camera was approaching the agent in comparison with when the still camera was used. A one-way ANOVA with the single factor of Condition (still, zoom, dolly, steadycam) showed a significant main effect, F(3, 48) = 13.54, p < .001. Post hoc comparisons showed significant differences between still and all other conditions (p < .001).
Similarly, results of the rating task for Question 2 (see Figure 8B) showed that participants felt more like the actor, that is, like being in the position of the actor in the scene when the camera was approaching the agent in comparison with the still camera. A one-way ANOVA with the single factor of Condition (still, zoom, dolly, steadycam) showed a significant main effect, F(3, 48) = 11.29, p < .001. Post hoc comparisons showed significant differences between still and all other conditions (p < .001).
Results of the rating task for Question 3 (see Figure 8C) showed that participants felt the zoom as less effective in making them feel like they themselves were approaching the scene.
A one-way ANOVA with the single factor of Condition (zoom, dolly, steadycam) showed a significant main effect, F(2, 32) = 6.77, p < .01. Post hoc comparisons showed significant differences between zoom and all other conditions (p < .01).
Results of the rating task for Question 4 showed that participants did not feel any difference among conditions in terms of the way they felt at ease while watching the video clips. A one-way ANOVA with the single factor of Condition (still, zoom, dolly, steadycam) showed no significant main effect, F(3, 48) = 0.59, p > .6.
Results of the rating task for Question 5 (see Figure 8D) showed that participants found the camera movement more realistic when the steadycam was used. A one-way ANOVA with the single factor of Condition (zoom, dolly, steadycam) showed a significant main effect, F(2, 32) = 6.91, p < .01. Post hoc comparisons showed significant differences between steadycam and zoom (p < .01) and between steadycam and dolly (p < .05).
The results for Question 6 (see Figure 8E) showed that participants found the camera movement more resembling a person's movement approaching the scene when the steadycam was used. A one-way ANOVA with the single factor of Condition (zoom, dolly, steadycam) showed a significant main effect, F(2, 32) = 16.14, p < .001. Post hoc comparisons showed significant differences between zoom and steadycam as well as between dolly and steadycam (p < .001).
Previous studies showed that during the execution and the observation of goal-directed actions rolandic mu rhythm shows ERD in both of its supposed components: central alpha frequency range and central lower beta frequency range (Perry et al., 2010; Leocani et al., 1997; Pfurtscheller & Lopes da Silva, 1999; Stancak & Pfurtscheller, 1996; Toro et al., 1994; Derambure et al., 1993; Pfurtscheller & Berghold, 1989; Pfurtscheller & Aranibar, 1979). In this study, we specifically investigated if this typical ERD/ERS pattern is modulated by the observation of video clips of hand actions filmed by (a) a camera moving toward the scene (possibly simulating the observer approaching the scene) and (b) the specific mode used for this approaching movement (zoom vs. dolly vs. steadycam).
Analysis of the present EEG data showed the following:
1. As shown in Figures 4 and 5, execution and observation of goal-directed hand movements produced significant ERD in both hemispheres for selected central alpha frequency ranges (8–14 Hz) as well as central lower beta frequency ranges (14–20 Hz) when comparing the mean value of frequency power of the whole time of action observation. No significant ERD for middle and high beta ranges (24–30 Hz) was detected. Furthermore, central alpha ERD was strongest during the participants' own hand action execution, whereas for lower beta action execution did not evoke stronger ERD than during action observation. In contrast, the steadycam condition evoked significantly stronger lower beta ERD than action execution.
These findings corroborate previous research describing central alpha and lower beta ERD during the execution as well as the observation of goal-directed hand actions. However, they also hint at some important differences between the different components of the rolandic mu rhythm. Our results suggest that ERD triggered by action execution is stronger within the central alpha frequency range than within the lower beta frequency range. In contrast, differences among observation conditions seem to be more strongly reflected in the beta frequency range. Previous studies found different EEG topographies for alpha and beta components of the mu rhythm (McFarland, Miner, Vaughan, & Wolpaw, 2000; Pfurtscheller, Pregenzer, & Neuper, 1994). It was therefore hypothesized that different neural networks are involved in the generation of these rhythms. Nevertheless, the differences found in our study to our knowledge have not been observed before and clearly need further investigation.
2. Remarkably, as shown in Figures 6 and 7, in both frequency ranges that had been described to show ERD during execution and observation of goal-directed actions, two common characteristics of the EEG time course were observed. First, during the time of video observation (Time Windows 1 and 2), descriptive analysis showed the strongest ERD for the steadycam, followed by the dolly, then the zoom, then the still condition. Statistically, in the alpha frequency range, significant differences were found between still versus steadycam as well as between zoom versus steadycam, and still versus dolly from 1000 msec after stimulus onset until stimulus offset. In lower beta frequency range, significant differences to steadycam included still versus steadycam, zoom versus steadycam, as well as dolly versus steadycam. These differences reached significance from the very beginning of stimulus presentation. Furthermore, in the second time window, also in the beta range, the difference between still and dolly became significant.
Second, during the rebound phase (third time window), in both ranges descriptive analysis showed the steadycam condition having the lowest power, but now first followed by still, then dolly, then zoom condition. In the alpha range, significant differences occurred between zoom and still and between zoom and steadycam, whereas in the lower beta frequency range, differences between zoom steadycam and dolly and steadycam were significant.
Taken together, these results indicate that reducing the distance between observer and observed agent, realized by moving the camera toward the scene, evoke stronger ERD of the mu rhythm during the observation of goal-directed hand actions. This difference was most pronounced when the camera movement was realized by using the steadycam (significantly different from the zoom and the still in the second and third window in alpha and from all other conditions in the first three windows in beta, except for the difference from the still in the third time window). Results of occipital control recordings did not support the interpretation that this difference might be because of increased overall attention evoked by the observation of these specific filmed actions.
As a note of caution, one should mention that the comparison between alpha and beta activity over central sites and occipital sites with reference to the mu rhythm might be affected by the reference choice—because using Cz as reference could attenuate the signal coming from electrodes surrounding the reference electrode. However, because the main analysis in this study was to compare EEG activity across conditions within each participant, the choice of reference should not influence relative increases or decreases in frequency synchronization.
Results of our rating task, as shown in Figure 8, show that participants clearly rated those movies in which the camera approached the scene as more involving than those filmed by a still camera. Furthermore, they perceived the movements of the steadycam as being the most natural and most resembling the movements of an approaching observer, thus eliciting the feeling that the observer him/herself would walk toward the scene.
Previous research showed that motor cortex activation during action perception is weaker if the observed action is less familiar to the observer (see Aglioti et al., 2008; Calvo-Merino et al., 2005, 2006). We suggest that our results might be explained by a similar effect regarding the perceptual familiarity of the observer induced by the type of camera movement used to film the scene. Our results indicate that a movement of the camera visible in the movie enhances observers' mirror mechanism. This could be related to a stronger feeling of involvement in the scene because of the approaching movement itself, as shown by results of the rating task, thus enforcing the mirror mechanism. It could also be because of the fact that a video recorded by a moving camera offers more depth cues and therefore more closely resembles real-life vision. Further research is necessary to clarify these issues. In any case, the effect only appears if the perceptual experience induced by the video clips and the visual experience we normally employ while moving ourselves actually resemble each other. Such similarity seems to depend on the filming technique and appears to be strongest when video clips are filmed with the steadycam.
This interpretation is also supported by our findings regarding the rebound phase (third time window). Here, the differences between still and steadycam are no longer significant in either band range, whereas there are significant differences between still versus zoom as well as between steadycam versus zoom and steadycam versus dolly. It seems as if subsequent to the observation of hand actions filmed with either still or steadycam, the alpha and lower beta frequency ranges showed weaker rebound than that measured for the dolly and especially the zoom condition. Interestingly, Koelewijn, van Schie, Bekkering, Oostenveld, and Jensen (2008) reported that beta oscillations were modulated in a similar way by the correct/incorrect nature of the observed action. In their study, they recorded EEG during an execution/observation task. In the execution task participants were asked to give button responses according to instructional cues. In the observation task, they saw other persons performing the same task, giving correct or incorrect responses. Results showed that beta oscillations during action observation were more strongly modulated if the action observed was “incorrect” according to the given cue. This stronger modulation was especially visible in the rebound phase, with beta showing a significantly stronger rebound when the answer was incorrect. We suggest that the results of our study might be interpreted as showing a similar effect if we consider the zoom and dolly conditions as presenting the observer with an “incorrect” or unnatural representation of the scene in so far as the movement of the camera does not resemble the movement of an actual person. Hence, the still and steadycam conditions would correspond to natural, thus “correct,” visual experiences.
In conclusion, we propose the existence of a familiarity effect regarding visual traces of camera movements in filmed stimuli modulating the mirror mechanism activation during observation of goal-related hand actions. That is, among videos dynamically reducing the distance between the observer and the observed agent, only videos simulating the “natural” vision of a human observer approaching an agent can elicit a significantly stronger ERD in comparison with videos showing the same scene from a fixed distance. Furthermore, the artificiality of other ways of simulating the dynamic distance reduction (such as zoom or dolly) might be reflected in differences in the time course of the rebound phase. This shows that the time course of mu rhythm ERD/ERS/rebound is modulated by the resemblance between the effect of camera movements and ordinary human vision. Familiarity with the visual experience provided by the video predicts mu ERD/ERS/rebound time course.
These findings shed new light on the neurophysiological mechanisms underpinning action execution–perception links. They indicate that dynamically reducing the distance to an observed scene enhances mirror mechanism. Furthermore, as this difference seems to depend on the use of the steadycam, producing a visual experience closest to natural human vision during movement, the study addresses the concern that many experimental designs are too artificial regarding perceptual processes in real life. The results of this study demonstrate the utility of ecologically designed studies for exploring action execution–perception links as well as further studies about the precise differences found here. Questions for future studies include the following: Can it be further explored how and why approaching a scene can enhance action–perception links? On which contextual features is familiarity with a stimulus dependent upon? And how do movement and mode of movement interact? Related studies are necessary to broaden our knowledge of social cognition and the function played in it by the mirror mechanisms, as well as for determining the perceptual basis of film experience (see Gallese & Guerra, 2012, in press).
This work was supported by the EU grant Towards an Embodied Science of Intersubjectivity (TESIS, FP7-PEOPLE-2010-ITN, 264828) to K. H. and V. G. The authors wish to thank Chiara Sebastiano, John McGraw, and the crew of la12studio for their most valuable help.
Reprint requests should be sent to Katrin Heimann or Vittorio Gallese, Department of Neuroscience, University of Parma, Via Volturno 39, Parma, Italy, 43120, or via e-mail: email@example.com, firstname.lastname@example.org.