Have I grooved to this before?

Learning a new motor skill typically requires the ability to convert actions observed from a third-person perspective into fluid motor commands executed from a first-person perspective. In the present study, we test the hypothesis that during motor learning, the ability to discriminate between actions that have been observed and actions that have been executed is associated with learning aptitude, as assessed by a general measure of physical performance. Using a multi-day dance-training paradigm with a group of dance-naïve participants, we investigated whether actions that had been regularly observed could be discriminated from similar actions that had been physically practised over the course of three days or a further set of similar actions that remained untrained. Training gains and performance scores at test were correlated with participants’ ability to discriminate between observed and practised actions, suggesting that an individual’s ability to differentiate between visual versus visuomotor action encoding is likely associated with general motor learning.

When learning a new motor skill, observing a model can facilitate the acquisition of complex new movement patterns -such as those required for sport, dance, or playing a musical instrument. Although numerous studies directly attribute gains in motor performance to physical practice (Lee, Swinnen, and Serrien, 1994;Savion-Lemieux and Penhune, 2004;Walker, Brakefield, Morgan, Hobson, and Stickgold, 2002;Wulf and Schmidt 1997), other studies indicate that some aspects of motor information can be learnt by observing a model before any physical attempts have been made (Blandin, Lhuisset, and Proteau, 1999;Bandura, 1985, 1987;Hodges, Williams, Hayes, and Breslin, 2007;Horn, Williams, and Scott, 2002). However, few studies have addressed whether an increased ability to retain the visual profile of observed movements is associated with a similarly increased ability to perform these movements following physical or observational experience. For instance, individuals who retain detailed visuospatial information regarding observed movements (e.g., placement of limbs in time and space, the physical relationship between different limbs, the timing and rhythm of movements) may be better able to access this information during subsequent attempts to perform these actions, thus leading to superior performance abilities. Alternatively, the level of detail with which a visually experienced action is encoded in long term memory may be unrelated to motor learning and performance ability if an individual is unable to adapt this information into corresponding motor commands. If the former scenario is supported by empirical evidence, measures addressing an individual's ability to retain movement information acquired through observation might provide a vital index of how well this individual could learn to perform complex new movements in new learning scenarios. In addition, if this relationship between action memory and performance aptitude is borne out, tests of action memory could be used to differentiate between individuals who learn actions best through observational experience, physical experience, or a combination of both in order to cater to individual learning needs.
Leading theoretical accounts of how we make sense of other people moving around us in a social world suggest that action understanding is achieved by a sensorimotor resonance process whereby observed actions are mapped onto corresponding components of an observer's existing motor repertoire (Gallese, 2003;Gallese, Keysers & Rizzolatti, 2004;Rizzolatti, Fogassi, & Gallese, 2001). In general, this correspondence between perception and action has been linked to action understanding as well as action learning (Buccino et al., 2004;Catmur, Walsh, and Heyes, 2007). Meta-analyses of action observation studies using neuroimaging document common regions of premotor and parietal cortices that are active during action observation as well as action execution (Grèzes and Decety, 2001;Caspers, Zilles, Laird, and Eickhoff, 2010). These overlapping regions may contribute to the formation of action memories by integrating kinematic and visuospatial information learnt through observation as well as execution.
Studies that report observational learning of novel movement patterns in the absence of concurrent physical practice demonstrate that sensory feedback is not essential for learning certain aspects of new movement profiles (Black and Wright, 2000;Maslovat, Hodges, Krigolson, and Handy, 2010;Kohl and Shea, 1992). In a task requiring participants to trace dynamic patterns using a computer mouse, observing another learner led to improvements in a subject's own movement trajectories, even without prior or concurrent physical practice (Hayes, Elliott, and Bennett, 2013). Specifically, using a between-subjects design, these authors demonstrated that the observation group improved between pre-and post-test when these participants were yoked to participants in a physical practice group, indicating that motor information regarding the intended tracing motions could be acquired through observation alone. The value of observational experience on subsequent motor performance has also been demonstrated using paradigms that require participants to perform immediately following observation as well. Mattar and Gribble (2005) found that participants who observed videos of individuals learning to manipulate a robotic arm were themselves able to immediately manipulate the arm better than control participants who had no prior observational experience. Additionally, performance accuracy was improved if the direction of force generated by the robotic arm (clockwise or counter-clockwise) in the execution condition matched the force-field seen during observation. In contrast, observing manipulations of the robotic arm in an opposite direction to the field encountered during execution led to poorer execution compared to receiving no observational experience, indicating that observational experience inconsistent with what is expected during physical performance can also reduce subsequent performance.
Collectively, these studies suggest that observational experience can engage the motor system in a manner that can either facilitate or attenuate performance gains across a variety of physical tasks, depending on the contextual congruency between observation and execution.
Evidence for the neurophysiological substrates that could support physical performance gains stemming from observational experience come from studies demonstrating common regions of cortical activity engaged when participants view actions that have been previously observed or executed (Calvo-Merino, Grèzes, Glazer, Passingham, and Haggard, 2006;Cross, Kraemer, Hamilton, Kelley, and Grafton, 2009). In a study that investigated the effects of a week-long dance-training intervention on action performance and perception, Cross and colleagues (2009) found that activity in premotor and parietal regions while observing dance movements was linked to the prior training context of each movement.
Specifically, both physically practised and passively observed movements evoked premotor and parietal cortices to a greater degree than untrained movements during action observation.
Since engagement of premotor and parietal cortices is frequently associated with visuomotor learning (Jonides, Smith, Koeppe, Awh, Minoshima, and Mintun, 1993;Binkofski, Buccino, Stephan, Rizzolatti, Seitz, and Freund, 1999), Cross and colleagues (2009) suggest that engagement of these regions when viewing actions that had been passively observed reflects their involvement in learning, even when no concurrent motor practice was present. In contrast to the findings reported by Cross and colleagues (2009) Findings from a recent dance-training paradigm similar to that used by Cross et al. (2009) add weight to the notion that the manner in which actions are experienced shapes their subsequent perception. In this study, auditory experience alone (i.e., listening to the soundtrack that could be paired with a dance sequence) was associated with weak engagement of premotor and parietal brain regions following training, while additional layering of visual and physical experience led to marked increases in activation within the same cortical regions . The increased neural response for each additional sensory modality was interpreted as evidence for increasing action embodiment as a consequence of multi-modal action experience during learning. The fact that physical experience was associated with the strongest engagement of parietal and premotor brain regions may be unsurprising, given that physical experience is consistently linked to greater performance gains relative to observational experience alone (Black and Wright, 2000;Maslovat, Hodges, Krigolson, and Handy, 2010;Cross, Kraemer, Hamilton, Kelley, and Grafton, 2009). These results may be due to the fact that direct, physical engagement of In support of this notion, other studies have demonstrated the aspects of performance that are least served through observational practice compared to physical practice. In a study involving a serial reaction time task, observational practice of key sequences led to poorer intermanual transfer, since an intermanual version of a sequence bears limited visual similarity to the observed model (Osman, Bird, and Heyes, 2005). In a separate study, Bird and Heyes (2005) found that observational practice of a tapped finger sequence was effector dependent, given that sequence production with untrained digits led to poorer performance. All together, these findings suggest that in order to benefit most from observational training, a model must demonstrate the task in a manner that is visually compatible with how the observer might reproduce the movement.
In order to accurately translate observed movements into motor commands, an observer must differentiate between his or her own physically executed movements and those executed by a model. One's ability to discriminate differences between observed and performed actions on the basis of differences in sensorimotor engagement could be intricately linked with overall performance ability -a relationship that, to our knowledge, has not yet been empirically examined. We hypothesised that dance-naïve participants who showed the best performance ability after a week of observational and physical practice with previously novel dance movements would also be better at discriminating between observed, practised, and untrained dance actions within a training-modality categorisation task. Such a pattern of findings would suggest that aptitude with learning to physically execute coordinated, whole-body movements is also associated with heightened abilities to encode and recall visuomotor experience specific to individual movements. The establishment of such a relationship could lead to the development of metrics that assess individual skill in sensorimotor differentiation, which could in turn be useful in classifying individual movement learning aptitudes.

Participants
Thirty participants with no prior history of dance training or experience with dance-based video games took part in this study. All protocols were approved by the Bangor University School of Psychology Research Ethics committee. All participants taking part in the study provided written informed consent before beginning any experimental procedures and were reimbursed for their participation. The final sample comprised 16 females and 14 males, with a mean age of 20.93 years (SD = 2.80 years). Performance test for four observed sequences, four practiced sequences and four untrained sequences on Day 4. C) Modality categorisation task depicting a still frame of the silhouette dancer from an example movement clip.

Stimuli and Apparatus
Twelve dance sequences were selected from Dance Central 2 (Harmonix Music Systems, 2011), a motion-capture based video game available on Microsoft's Xbox Kinect™ 360 console (see Figure 1). These sequences were selected on the basis of gender-neutral choreography and minimal background graphics, after which they were randomly assigned to three separate groups. Game choreography was set to popular dance music, with an average song length of 2 minutes and 19 seconds (SD: 12 seconds), and an average tempo of 115 beats per minute (SD: 10.24 bpm). For each participant, these sequence groups were counterbalanced across three training conditions: physically practised, observed, and untrained sequences. Each sequence group did not significantly differ in difficulty rating, duration, or beats per minute. All participants experienced all sequences from the three training groups.

Behavioural training procedure
During physical practice, participants performed four dance sequences approximately 2 metres in front of a wall-mounted Sharp 52-inch flat-screen TV. The Kinect™ motion capture system was calibrated so that 3D full body motion for each participant was Performance was scored online using the videogame's Kinect motion-capture hardware, which matches the overall silhouette of the performer with the silhouette of the computer avatar. During the execution of a movement, a performer's silhouette must closely mirror the template of the avatar in order to obtain a high score. The game generates a total performance score based on mirroring accuracy after each sequence is completed, whereby higher scores indicate better mirroring of imitated moves. Scores obtained for each sequence were averaged to reflect overall performance ability on each training day. Importantly, in the native gameplay context, real-time feedback appears on the right side of the screen as participants perform each dance movement. This feedback includes a dynamic silhouette of themselves dancing, as well as verbal feedback letting them know how well they were matching their movements to the avatar on screen (terms such as 'flawless!!' or 'almost!!' appear) and numeric feedback (a participant's overall score tally grows depending on performance. In order to keep the physical training condition as similar as possible to the observational condition, the part of the screen where real-time feedback was displayed was covered so that participants could not see how well they were doing as they performed (however, participants could catch a glimpse of their final dance score as it was briefly flashed up in the centre of the screen after each performance, as there was no way to hide this feedback or disable this feature in the game).
For the observation condition, participants were seated in front of a computer running Psychophysics Toolbox 3 in MATLAB R2013a (Mathworks, Inc.) and observed four dance sequences recorded from the video game (see Figure 1B). After watching each sequence, participants were shown 8 movement clips and asked to state whether each movement had been featured in the preceding sequence ("Did you see this movement in the video you just watched?"). Participants responded "yes" or "no" using the computer's cursor to select the respective option. Half of the displayed movements were extracted from the preceding sequence, while the other half were extracted from sequences not used during training. Total accuracy for each day was calculated as the number of movements correctly identified from the observed sequences.
For three consecutive days, participants experienced both physical practise and observation conditions. On the fourth day (test), participants physically performed all sequences featured in physical and observational training, in addition to four untrained sequences. To limit the impact of instructional differences between physical practice and observational experience on test performance, participants were only made aware of the test phase on the last day of participation. In addition, they were never explicitly instructed to try to learn or memorise the sequences they experienced during physical or observational training conditions (see Grèzes, Costes & Decety (1999) and Badets,

Blandin & Shea (2006) for further discussion of the impact of instructions on learning).
The four scores generated for each condition were averaged to reflect a global measure of performance ability for each training condition at test.

Training modality categorisation task
After completing the dance task on the fourth (test) day of the study, participants then completed a brief computer task requiring them to categorise movements into physically practised, observed, and untrained movement conditions (see Figure 1C). The stimuli for this task featured moving body silhouettes performing the individual choreographed moves that composed the longer movement sequences used in the game. Critically, these stimuli were devoid of the complex and dynamic background graphics and associated music present during training, which could be used to recognise actions from their respective training contexts using visual and auditory cues specific to the videogame. For a comparison of movements as presented during training compared to their presentation during scanning and categorisation, see Figure 2. Each move performed by the silhouetted dancer was approximately 2 seconds long, and was presented once followed by the question "In what context did you see this movement?".
Participants were required to select the appropriate training condition ("physical", "observed", or "untrained") using the computer's cursor. Total accuracy was calculated as a percentage of all correctly categorised movements. Accuracy scores for each training condition were also generated for the purposes of analysis.

Design
To assess whether performance ability for physically practised movements improved across training, a repeated-measures ANOVA was used to compare performance scores for physically practised movements across all three days of physical training. A repeatedmeasures, within-subjects ANOVA was similarly conducted to determine whether accuracy on the observation task also improved across training prior to test. Modalityspecific performance at test was examined using a one-way, within subjects ANOVA by comparing average differences in performance between physically practised, observed, and untrained movement sequences. Differences in categorisation ability between the three groups of movements were also examined using a one-way, within subjects ANOVA. Post-hoc pairwise comparisons were conducted using Tukey's HSD. Performance scores were then correlated with modality categorisation to examine the association between specific sensorimotor action memory and performance ability.  Results from the within-subjects, repeated measures ANOVA indicate a main effect of day (see Figure 4), F(2, 58) = 3.73, p = 0.03, ηp 2 = 0.11, indicating that participants became increasingly accurate at identifying whether or not specific movements were present in the observed sequences. Pairwise comparisons revealed that performance accuracy was significantly higher on Day 3 than Day 1. Overall differences in scores across the three days of training can be described as a linear trend, F(1, 29) = 6.43, p = 0.02, ηp 2 = 0.18.

Figure 5. Test day performance scores for all sequences.
Error bars indicate standard error of the mean. *significant at p < 0.05.
A main effect of training modality was observed for performance scores during the test session on day 4, F(2, 87) = 3.92, p < 0.05. Post-hoc comparisons (Tukey's HSD) revealed that physically practised sequences were performed significantly better than untrained sequences (see Figure 5). Pairwise comparisons revealed that performance on observed sequences did not significantly differ from physically practised or untrained sequences.

Training modality categorisation task
At test, participants were able to recall the correct training modality for viewed action silhouettes at a rate well above chance (mean recall rate = 73.17%; chance rate = 33.33%see Figure 6). Given that the assumption of homogeneity of variance was violated, the Brown-Forsythe Fratio is reported. A main effect of modality was observed, whereby untrained movements were categorised more accurately compared to physically practised or observed movements, F(2, 71.05) = 10.09, p < 0.001. The rate of categorisation was highest for untrained movements (84.17%), followed by physically practised (71.67%) and observed actions (63.67%). Pairwise comparisons indicate that accuracy for untrained movements was significantly higher than accuracy for physically practised movements and accuracy for observed movements. Accuracy for physically practised movements and observed movements did not reliably differ (see Figure 6).

Figure 7. Categorisation versus performance.
Test day modality categorisation accuracy correlated with test day average performance scores.
As predicted, overall accuracy rates for modality categorisation correlated with global performance scores on test day, r(30) = 0.60, p < 0.001, indicating that participants who overall performed dance sequences better at test also scored higher on modality categorisation ability (see Figure 7). Accuracy for categorising physically practised as well as observed movements was respectively associated with performance ability for physically practised, r(30) = 0.43, p = 0.02, and observed sequences, r(30) = 0.40, p = 0.03, indicating that performance in the modality categorisation task was associated with performance ability for observed and practised sequences, but not for untrained sequences (see Figure 8).

Discussion
The primary aim of this study was to investigate whether an individual's ability to learn novel, whole-body actions is associated with his or her ability to discriminate the learning context of these actions following training. A second question we explored is whether an individual with a heightened ability to recognise learning modality-based differences in action sequences might also be capable of reproducing these movements with greater accuracy following training. As hypothesised, our findings demonstrate a positive association between participants' ability to accurately classify learning modality and their post-training dance performance, indicating that increased performance aptitude does indeed track with the ability to recall the sensory modality through which an action was originally learned.
In order to probe action learning, participants in this experiment were asked to identify the training context in which each action was originally learned using novel versions of the action stimuli, stripped of the rich visual and auditory cues that might make linking specific actions to their original training contexts easier. These stimuli were used so that participants had to rely more on prior sensorimotor experience to identify training contexts, rather than the rich visual and auditory cues specific to the videogame that they experienced during physical and observational training (see Figure 2). Across all training contexts (physically practised, observed and untrained), accurate categorisation of these actions was significantly above chance, indicating that the type of sensorimotor experience associated with an action (or lack thereof) was reliably recalled. Within this task, response accuracy did not significantly differ between physically practised and observed actions, although higher categorisation accuracy for both forms of experience was associated with an increased ability to perform these trained actions. These results suggest that participants' learning was also associated with their ability to discriminate between visually encoded versus physically experienced actions.
In contrast, categorisation of untrained actions was not associated with performance aptitude for untrained actions, despite categorisation accuracy being highest for movements from this set. It is plausible that the high categorisation accuracy for movements from this training category is due to the novelty of these previously unseen/undanced actions. has shown as well in a recent study using the same dance video game set up as that used in the present study , explicitly requesting participants to try observe with the intention to learn increases observational learning compared to passive learning alone, and relative to untrained actions. It is thus perhaps all the more striking that participants' classification accuracy for physically practiced and observed sequences is statistically indistinguishable, and tracks with their ability to perform these sequences. A challenge for future work will be to more closely examine how the accuracy with which an action is executed relates to the accuracy with which an individual encodes visuospatial and kinematic features during action experience, as well as to examine how the varying instructions about learning intentions further shapes this relationship.
In general, humans appear to be proficient at differentiating previously observed action profiles. In a study by Urgolites and Wood (2013) that investigated visual action memory, participants observed a series of computer-animated actions performed by an avatar (such as jumps, arm raises, and crouches) and were then presented with pairs of actions featuring a previously seen and an unseen action. For actions observed between one and five times, accuracy for selecting seen over unseen actions ranged between 76% and 81%, suggesting that visual properties of observed actions can be accurately recalled from long-term memory.
The authors conclude that acquisition of new actions may critically depend on integrating new sensorimotor information with pre-existing action templates held within long-term memory. The increased engagement of sensorimotor brain regions documented while dancers observed previously practised actions in Calvo-Merino and colleagues' study (2006) could reflect this type of long-term action memory facilitated by physical experience. For actions that have only been observed for a similar period of time, lesser engagement of sensorimotor regions when observing these actions may be indicative of reduced sensorimotor integration.
In essence, frequently observed actions that are never accompanied by physical practice may be encoded primarily using visual information, while actions that are observed as well as practised benefit from both visual as well as motor encoding (Calvo-Merino et al., 2006). The additive impact of these two forms of encoding could promote the retention of physically practised actions in long term memory, given that performers are able to recall and perform routines trained many years ago (Stevens, Ginsborg, and Lester, 2011). In contrast, extensive visual experience with actions in the absence of physical practice may not facilitate gains in performance to an equal degree, despite facilitating detailed visual encoding.
As the present study demonstrates that the ability to recognise observed and physically practised actions is linked to behavioural performance aptitude, an area of interest for future work could be to examine the degree to which visuomotor representations are separable at a neural level. If individuals who are poor at recalling the original training context of actions nevertheless show distinctions in neural engagement when observing these movements, this discrepancy would suggest that neural differentiation between observed and physically practised movements does not necessarily translate into direct awareness or memory of training context. However, if neural differentiation between learning modalities was predictive of later performance gains, this activation could provide an index of how much an individual might learn through an observational training paradigm, even if he or she has difficulties accessing modality information at an explicit level of awareness. Such metrics could then be used to devise appropriate training interventions depending on individual learning profiles. In addition, stimuli used to probe differences in movement encoding could be further reduced to minimal motion cues in order to gauge whether practicerelated information can be conveyed in the absence of cues to human form. If participants were able to classify training related differences using simplified movement stimuli, this would point to action encoding mechanisms that are not necessarily reliant on detailed human models, widening the scope of visual instructions that could be used for new action training. More broadly, the approach and findings of the present study hold potential value for specific motor-training paradigms by demonstrating how individual differences in movement encoding might be linked to motor learning and performance.