Human beings can learn to perform a variety of skilled movements ranging from dancing, painting, playing musical instruments, riding a bicycle, driving a car, playing a wide range of sports and so on. This ability to perform such complex movements is undoubtedly one of the characteristic functions of the human brain which enables us to interact with our environment. Even simpler actions such as reaching for and grasping a cup of tea, holding a pen for writing and tying one’s shoelaces are complex problems for the nervous system, but are performed in a rather seamless manner by us on a day-to-day basis. Motor skill learning is a key ability that enables us to acquire and store such actions and distinguishes humans from other species. As a society we value this ability which enables us to make a wide variety of movements, and are awed by the fine skills of musicians, dancers and sports players. Scientists have long been fascinated by this capacity of humans to learn complex skilled movements and have tried to probe its underlying neurophysiological basis to advance fundamental neuroscientific understanding as well as for translational goals.

Literature available on skill learning can sometime confuse readers and/or young researchers getting into this field of study. As a master’s student doing her thesis on skill learning and retention in 2014, the now co-author of this paper (Goldy Yadav) often came across research papers loosely using the term “skill” for tasks assessing motor performance under controlled conditions (such as motor learning under externally induced perturbations). Despite several decades of work, we still lack a clear understanding of what we mean by skill in human motor learning research field. Challenges in formulating a formal definition of skill has been felt by many, as our views are fragmented and lack consensus (nicely elucidated more recently by Christensen, 2019). This can have serious implication for result interpretations and replication of findings in this area of research. Moreover, this can be very misleading not only for fundamental motor skill learning research, but also motor rehabilitation programs which aim to translate lab-based research findings for patient benefits. A much-needed consensus on definitions and measurement standards of task paradigms is critical for reducing the time lag in translation of scientific discoveries for clinical practices (Shumway-Cook and Woollacott, 2007; Morris et al., 2011).

This review paper is therefore an attempt to obtain clarity on what is meant by “skill” by highlighting key concepts that came out of some seminal work and how those ideas can be used as a strong framework for current and future skill learning research. We aim to shed light on how our understanding of skill in the context of human motor learning has evolved with time. First, we lay out what is broadly meant by motor learning and highlight early theories of motor learning that shaped our understanding of skills. Next, we delve deeper into a specific form of motor learning that is of interest to us, i.e., motor skill learning and attempt to understand the specific features of skill based on prevailing research in the field. Finally, we conclude by emphasizing the key features of skill that should be incorporated in motor skill learning research work to better assess, measure and understand the underlying nature of human skill behavior. We also provide some examples of existing model tasks that very well incorporate these features.

Motor learning can be broadly defined as practice induced changes in motor performance. The early era of motor learning research involved studying motor behavior using a “task oriented” approach and the term “skill” was often synonymous with perceptual-motor performance. At that time there was little emphasis on variables or factors that underlie motor learning at an individual level. Such work explored global features of motor learning in humans (Hull, 1943), and often involved gross measurements and scoring of movement parameters such as movement time. Over time, researchers started shifting from these global measures to local measures in order to better understand the nuances of human motor behavior. Some of these assessments involved studying the amount of time taken to process visual information before initiating movements, the role of attention in motor performance and how error detection occurs during the movement (Schmidt, 1975). As a result, a number of theories were proposed to explain human motor learning. For instance, Keele (1968) proposed that “motor programs” are required for making precise and accurate bodily movements. These motor programs were thought of as a sequence of “stored commands” in the brain which enables us to make a set of movements by seamlessly incorporating the external feedback. Later Adams (1971) argued that motor programs are abstract memory forms that are prepared before movement initiation and contain information about the pattern of muscle contraction and relaxation for a given movement type. In other words, the human brain acquires and stores a set of stored muscle commands that are ready to use at any given time for making a movement.

Around the same time period, another popular theory about motor learning was proposed by Schmidt (1975)- this theory was an attempt to explain why we need not have distinct individual motor commands for each movement that we make. According to this theory, there are clustered and generalized motor programs called “schemas,” which can be fine-tuned to make specific discrete sets of movements. In fact, long ago Bartlett (1932) had suggested that such schemas are critical given the limited storage capacity of the human brain, and possibly underlie our astonishing ability to quickly learn novel sets of movements. Interestingly, the idea of schema formation underlying motor learning is evident in some recent work as well (Newell, 2003; Sherwood and Lee, 2003; Shea and Wulf, 2005; van Vugt et al., 2014; Willey and Liu, 2018). This notion has also been used to explain motor skill behavior. For instance, van Vugt et al. (2014) studied expert pianists and found that highly skilled complex actions emerge from a combination of simpler movements known as motor primitives (equivalent of motor schema). The authors argue that our nervous system efficiently uses these motor primitives to produce a wide repertoire of complex movements that can be enhanced through skill learning. We believe these concepts of skill can also have implications for studying and understanding expert motor skill behavior which typically operates on a different timescale than lab-based experimental setup requiring hundreds of hours of training, but may share underlying common features (for interesting work on elite performers see Wulf et al., 2002; Beilock and Carr, 2004; Wulf and Su, 2007; Abdollahipour et al., 2015; Singh and Wulf, 2022).

According to the schema model of motor learning (Schmidt, 1975), the nervous system requires four parameters for a goal-directed movement–(i) initial task conditions, (ii) response specifications of the movement to be made, (iii) sensory consequences of the executed movement, and (iv) outcome of the executed movement. Eventually after producing a set of movements, the nervous system extracts abstract information about the relationship among these four parameters. This abstract relationship is stored as schema to enable us to perform these movements later when required. Once a schema has developed, only two inputs/specifications are needed to this schema in order to execute a movement- (i) initial task conditions and, (ii) the desired outcome of the movement. Schmidt (1975) argued that since these inputs to the schema are never exactly the same, every movement that we execute is almost always novel, although appearing similar from a broader level. This view finds support from Bartlett (1932) who said, “we do not execute the movement exactly as we have made it before.” This may explain the inherent variability underlying complex movement execution (more details on variability in Section “3.3. Skill as performance optimization”). Basically, a motor response is nearly never repeated given the number of possibilities that may arise from a given schema. In this process of motor execution, sensory outcomes of the movement are anticipated using the schema and each of the resultant sensory outcomes are compared with respective incoming sensory information (interoceptive and proprioceptive feedback) during or after the movement. In case of a mismatch between the anticipated and actual sensory information, an error value is assigned which is later used to update the schema. The primary goal of the nervous system, therefore, during motor learning is to reduce such errors.

The errors resulting due to the mismatch between anticipated and actual sensory consequences can also serve as a substitute for knowledge of results (KR) which can be used to update the schema (Salmoni et al., 1984; Winstein, 1991; Guadagnoli and Kohl, 2001). KR is one of the key ways to improve motor performance and enhance learning. During a motor learning task, KR is mostly presented in three ways so as to facilitate corrective responses (Adams, 1987)–(i) by directly presenting the pattern of the motor response to the participant for which error must be inferred, (ii) feedback in the form of the actual motor response that the participant made along with the ideal expected motor response, where the difference between the two is the error which can be easily inferred by the participant; and (iii) partial error information for the motor response made by the participant. In addition to providing KR to improve motor performance, learning by observation is another effective way. Newell (1985) proposed that learning through observations may help in easy identification of errors, and hence enable efficient motor responses. Motor learning by observation has been mainly shown to enhance performance in a number of studies (Shea et al., 2000; Stefan et al., 2005; Wulf et al., 2005; Celnik et al., 2006; Granados and Wulf, 2007; Nishizawa and Kimura, 2017; Kawasaki et al., 2018; Jayasinghe, 2019), with some distinctive effects reported for observation and motor task type (please see Sorgente et al., 2022 for details). KR and observation, therefore, may play a critical role (as external factors/interventions) during skill acquisition to optimize skill performance (more details in section “3.3. Skill as performance optimization”).

The points described above highlight the broader aspects of human motor learning. Now let’s focus more specifically on how motor learning is studied in the laboratory. In this context it is largely investigated using two classic paradigms- motor skill learning and motor adaptation. Our understanding of each of these two paradigms has evolved with time, enabling us to better understand the nuances of human motor behavior. Motor skill learning is characterized by performance changes beyond baseline/starting levels in the absence of any external perturbation (more details in the next section). Such changes may be characterized, for instance, by increased spatial and temporal accuracy of movements over the practice session (Adams, 1987; Reis et al., 2009; van der Steen et al., 2014; Yadav and Mutha, 2020). Features of skill learning include, but are not limited to- reduction in trial-to-trial variability, changes in speed-accuracy relationship or other performance limiting variables, offline gains, acquisition of new control policies and exploration (Dayan and Cohen, 2011; Shmuelof et al., 2012; Telgen et al., 2014; Sternad, 2018; Vassiliadis et al., 2021; Du et al., 2022). In contrast, motor adaptation involves modification of motor output to account for the effects of externally induced perturbations and restoration of performance to pre-perturbation levels (Krakauer and Mazzoni, 2011; Morehead et al., 2017; Kumar et al., 2020). Figure 1 (Adapted from Chapter 8, Sternad et al., 2014) shows clear distinction between motor adaptation and motor skill learning. As evident from the figure, the goal during motor adaptation is to return to baseline performance (low motor error) by reducing the errors arising from external perturbations. On the other hand, skill learning involves gradual reduction of errors that are high in the beginning of the session because of the novel nature of the movements and the need to acquire a new control policy. Skill acquisition involves de novo learning in contrast to adaptation (Sternad, 2018; Krakauer et al., 2019).

In addition to these behavioral differences, adaptation and skill learning are also thought to be dependent on distinct neural systems. For instance, motor skill learning appears to be primarily mediated via primary motor cortex and basal ganglia circuits (Floyer-Lea and Matthews, 2005; Halsband and Lange, 2006; Doyon et al., 2009; Dayan and Cohen, 2011; Cantarero et al., 2013; Spampinato and Celnik, 2018) while adaptation seems to requires an intact cerebellum and posterior parietal cortex (Tseng et al., 2007; Mutha et al., 2011a,b; Morehead et al., 2017; Kumar et al., 2020). It is therefore critical to keep these differences between motor skill learning and adaptation in mind while formulating research questions on the nature of human motor behavior to avoid any ambiguity and over-generalization across paradigms (Ranganathan et al., 2021). Further, understanding these distinctions also help exercise caution during data interpretation.

In this paper, we focus on motor skill learning to unravel the characteristic features of skill. With this goal in mind, we review the prevailing ideas of what is meant by “skill” in this field of study. We are specifically interested in skill learning because it holds tremendous implications for motor rehabilitation- complex motor skill behavior in humans and its rehabilitation following neurological damage are critically dependent on the ability to learn and regain lost motor skills. We therefore believe that it is very crucial to have some consensus and clarity on what we mean by “skill.” Adopting clear working definition(s) of motor skills is important when our collective goal is to better understand human skill behavior.

The ability to execute and learn a variety of skilled movements is an astonishing human feat. Skill emerges as a consequence of learning/training/extensive practice. Azim (2014) said- “Few of our limb movements will ever gain immortality like Willie Mays, but what we accomplish every day is remarkable.” Study of human motor skills is of tremendous neuroscientific interest, and several attempts have been made to define motor skills over these years. Pear (1927) loosely defined skill as well-adjusted and integrated motor performance, which is dependent on learning and optimal motor output, and therefore distinct from a mere capacity to perform a given motor task. Later in 1952, Guthrie stated that motor skill is an ability to perform with high certainty that require minimum energy and time. Willingham (1998) proposed a neurophysiological theory of motor skill learning in which he describes that skill learning emerges out of motor control processes. While motor control involves planning and execution of movements (Hommel, 2009; Krakauer et al., 2019; Merel et al., 2019), motor skill learning, in addition to involving movement selection and execution, is the process through which movement quality is improved with practice (Adams, 1987; Willingham, 1998; Dayan and Cohen, 2011; Chen et al., 2018; Sternad, 2018).

A key form of human motor behavior is skill learning, which is distinct from another popular task paradigm known as motor adaptation. Characterizing key features of motor skills, we believe, is the first critical step toward understanding how humans acquire this quintessential ability through training. Based on prior work by other researchers outlined in this paper, it appears that motor skill can be broadly characterized by three key features (Figure 2)- (i) optimal movement selection and execution, (ii) improved movement speed and accuracy, and (iii) reduced movement variability and error, as a consequence of extensive practice. Here, the first feature may rely on external cues such as target information and thus primarily involve strategic/conscious motor control processes in the early stages of skill acquisition. The latter two, on the other hand, may mainly be driven by unconscious processes that help overcome task constraints resulting in a performance state beyond baseline level. A culmination of these aspects, we believe, finally results in a seamless automatic skill behavior that we observe in humans, which may further evolve as expert-level performances (with extensive practice/training) in the form of dancing, painting, playing musical instruments, sports, and so on.

The three key features mentioned above, therefore, should be included in any task designed to capture and study motor skills. For example, lab based experimental procedures of skill learning (wherein extensive practice spanning months and/or years is not possible) can exclusively assess individuals on motor tasks with a given speed-accuracy condition that requires them to acquire optimal movement (from a repertoire of possible movements) leading to low error/variability. Skill tasks that very well capture these aspects are, for instance, those that involve tracing movements within a given speed-accuracy criterion and pinch-based movements requiring optimal force production to make an accurate goal-directed movement. In addition to these, even point-to-point reaching movements that involve moving the effector end-points as well as the joints (elbow and shoulder for instance) in a manner that reduces movement-related errors toward a target (in the absence of any external perturbation) could be a simple skill task to use in the lab (especially for patient populations with motor disability). Such task setups and assessments can capture the most fundamental aspect of human motor skill behavior which involves selection and execution of movements that optimize time and energy (internally) while yielding successful purposeful outcomes (externally). Hence, systematic use of such tasks would ensure homogeneity at the methodological level, help build more comprehensive theoretical framework and speed up cumulative progress of human motor skill research. Further, the resultant clarity would help curb replication crisis in human motor behavior research and benefit translation of fundamental knowledge to clinical settings- both relying primarily on agreed upon measures and definitions for scientific knowledge enhancement and knowledge transfer.

Both authors listed have made a substantial, direct, and intellectual contribution to the work, and approved it for publication.