Distinct modes of executing movement sequences: Reacting, associating, and chunking
Highlights
► Movement sequences can be learned by two mechanisms. ► The sequencing mode is based on motor chunks. ► The associative mode is based on priming forthcoming reactions. ► The associative mode may be overshadowed by the chunking mode.
Introduction
The performance of sequential motor skills sometimes involves rapid execution of a movement series without much need for external guidance (like when typing in a familiar phone number, or playing by heart a highly familiar piece on a musical instrument). However, in other situations these skills seem to remain largely externally guided (like when playing a less familiar piece on a musical instrument from sheet music). Initial cognitive models assumed that movement patterns are guided entirely by the feedback of the preceding movement. These are known as associative chain or closed-loop models (e.g., Adams, 1971, James and James, 1890). Later models acknowledged that movement patterns eventually require less and less external guidance because internal, mental representations indicate the order of the individual movements (Keele, 1968, Lashley, 1951, Povel and Collard, 1982, Restle, 1970, Rosenbaum et al., 1983, Schmidt, 1975). These representations are often referred to as ‘motor programs’ and allow for internally guided (i.e., open-loop) control. More recent studies showed that these mental representations may involve different codes (egocentric and allocentric spatial, motor and even verbal). Which codes develop exactly is influenced by factors like the amount of practice (e.g., Hikosaka et al., 1999), the nature of the stimuli (Shea et al., 2011, Tubau et al., 2007), the length and the difficulty of the movement sequence, and the provision of feedback (e.g., Kovacs et al., 2010, Kovacs et al., 2009). Fast and smooth execution of movement sequences is probably dependent especially on representations using movement related (i.e., motor) codes (e.g., Shea et al., 2011).
The present study aimed at showing that the practice of short and discrete movement sequences yields not only an internally guided, but also an externally guided control mechanism. Distinguishing different control mechanisms in discrete sequencing skill is important for explaining flexibility in new situations (which involves transfer of skill), and for a basic understanding of the processes involved in discrete sequence skill and how these processes relate to well-known sequential skill models (e.g., Hikosaka et al., 1999, Keele et al., 2003). Moreover, disentangling different cognitive learning mechanisms provides implications for the study of the neural correlates of sequencing skill (e.g., Dayan and Cohen, 2011, Hikosaka et al., 1999). To assess the contribution of underlying control mechanisms we here specifically use keying sequences because the limited duration of individual key presses makes these sequences well-suited to study the underlying control processes (MacKay, 1982, Rhodes et al., 2004).
Studies of sequential keying skills have usually focused either on relatively short, fixed and discrete stimulus–response (S–R) sequences that prompt internally guided execution (such as for example in the Discrete Sequence Production or DSP task, e.g., Rhodes et al., 2004, Verwey, 1999), or on the more continuous cycling through a single, relatively long S–R sequence that encourages externally guided control (like in the Serial Reaction Time or SRT task, Abrahamse et al., 2010, Keele et al., 2003, Nissen and Bullemer, 1987). These two research orientations produced different models of sequencing skill. Inspired by the clear similarities between the DSP and SRT tasks (see below), the current study proposes – and supports – the idea that practice in the DSP task yields both open- and closed-loop sequencing skill but that a particular manipulation is required to reveal the closed-loop skill in that task.
In the DSP task, participants initially respond to fixed series of position stimuli by pressing the spatially corresponding key following presentation of each stimulus. So, initially the sequence is externally driven. These sequences are discrete in the sense that they involve fixed series of only 3 to 7 stimuli with considerable time between successive sequences. It is assumed that, with practice, execution in the DSP task becomes internally controlled due to the development of representations referred to as motor chunks (see e.g., Miller et al., 1960, Paillard, 1960). A motor chunk codes successive movements such that they can be selected, initiated and executed as if they constitute a single response (Verwey, 1996, Verwey, 1999). Motor chunking is characterized by a relatively slow first key-press (caused by identifying the stimulus, selecting or assembling a motor chunk, and initiating the sequence), followed by a series of key-presses that can be executed fast and smoothly because the motor chunk provides movement related information. Other indications for the relatively automatic execution of motor chunk use are that DSP sequences can be executed without key-specific stimuli past the first (Verwey, 1999, Verwey, 2010), and that a secondary task has only little effect on the execution of a DSP sequence (Verwey, Abrahamse, & De Kleine, 2010).
With respect to the execution of discrete keying sequences, Verwey (2003b) distinguished between execution in a reaction mode and in a chunking mode.1 In the reaction mode participants carry out a keying sequence by translating each key-specific stimulus into the appropriate response (e.g., Hikosaka et al., 1999). Such execution, which is typical in the case of unfamiliar sequences, requires continuous involvement of perceptual and (higher-)cognitive processing resources. With practice, the chunking mode develops. According to the dual processor model (Verwey, 2001) this mode involves two successive processing steps. First, a motor chunk is selected and loaded into a motor buffer by a cognitive processor before execution of the sequence (i.e., sequence preparation). Second, the key presses represented by the motor chunk are executed in rapid succession by a motor processor reading information from the motor buffer. This implies that perceptual and (higher-)cognitive processing resources are primarily used during sequence preparation (Verwey et al., 2010), and that sequence execution is under open-loop control.
In the typical SRT task participants also respond to each of a series of stimuli by pressing spatially corresponding keys. However, unlike the DSP task, an SRT task involves only one sequence which usually is longer than the DSP sequences (typically 12 as opposed to 2–7 elements). Furthermore, SRT sequences include no clear transition between successive instances of the sequence (i.e., they are not discrete), involve limited practice (less than 100 repetitions per sequence, as opposed to 500 in the DSP task), and participants are usually not informed that stimuli follow a particular order (in order to induce only incidental or implicit learning). Most of the time, the SRT task involves response-to-stimulus intervals (RSIs) in the range of 200 to 500 ms. Learning in the SRT task is indicated by a gradually faster execution with practice, and a sudden slowing when the stimulus order is changed.
SRT studies have led to the notion that participants benefit from practicing the sequence because associations develop between the representations used in successive responding. As recently argued by Abrahamse et al. (2010), these associations can occur at the perceptual level (e.g., Remillard, 2003), the response selection level (Schumacher and Schwarb, 2009, Schwarb and Schumacher, 2010) and/or the response level (e.g., response location learning; Willingham, Wells, Farrell, & Stemwedel, 2000).2 We propose to refer to this mode of sequence execution in which individual key-specific stimuli are identified, as the associative mode. In contrast to the performance in the chunking mode the associative mode shows relatively modest improvement (i.e., average response times do not typically drop below 250 ms), that is more or less similar for all key-presses, and sequence execution remains dependent on stimulus presentation and under closed-loop control. The latter is supported by, among others, Cohen and Poldrack (2008) who showed that even after considerable SRT practice participants are able to inhibit their response when an unexpected stop-signal is presented within the stream of sequentially ordered (go-)stimuli.
Whether motor chunks underlie sequence production in the SRT task has recently been a topic of debate (Jiménez, 2008, Jiménez et al., 2011). Motor chunks may be used in the SRT task in case of consistent temporal and spatial inhomogeneities in the stimulus presentation, like with left-to-right keying orders and reversals (Kirsch et al., 2010, Koch and Hoffmann, 2000), and when pauses are included at certain, fixed positions (Verwey, 1996, Verwey and Dronkert, 1996; also see Stadler, 1993). Still, Jiménez et al. (2011) provided support for the notion that chunking is not typical for SRT performance. They first used the color of stimuli to induce a uniform chunking pattern across participants in the training phase (i.e., different colors were used for groups of three stimuli). Indeed, response times of individual key presses suggested the use of motor chunks. However, when in the subsequent test phase the color coding was removed and the task was turned into a typical SRT task, these indications for chunk use disappeared again. This suggests that SRT sequences involve chunking only in very specific situations (see also Jiménez, 2008). Important in the present context is that removing the colors in the Jimenez et al. (2011) study did not entirely eliminate the benefit of the familiar over an unfamiliar SRT sequence, as if the use of motor chunks during practice had not prevented the development of associations between successive sequence elements.
Participants with an explicit, in-depth knowledge of the element order that they can verbalize are said to have structural knowledge whereas participants merely knowing that there is a regularity would have judgment knowledge (Dienes & Scott, 2005). The DSP and the SRT tasks have in common that a considerable number of the participants do not develop substantial structural sequence knowledge (Verwey, 2010, Verwey et al., 2010). Whereas this is not so surprising for the SRT task – which is actually designed to be an implicit learning task and may leave participants even without judgment knowledge – absence of structural knowledge is rather surprising in the context of the DSP task where sequences are short, discrete, and highly salient. In fact, in this task participants are usually told about the fixed element order. In both the DSP task (Verwey, 2010, Verwey et al., 2010, Verwey et al., 2009), and the SRT task (Curran and Keele, 1993, Liu et al., 2007, Willingham et al., 1989) participants with substantial structural knowledge are often a little faster than less aware participants, but skill in these two tasks typically does not depend much on explicit knowledge.
For the DSP task, it is assumed that the cognitive processor can probably also use explicit sequence knowledge triggering responses in parallel to the motor processor, but this contribution of the cognitive processor may well decrease as motor chunks develop and less time is available during sequence execution (Verwey et al., 2010). Moreover, explicit knowledge can be used to load a limited number of responses into the motor buffer (Verwey, 1996). In the SRT task, the gain of explicit knowledge is attributed to aware participants consciously anticipating stimuli (Willingham et al., 1989), which would be independent of the associative mechanism underlying implicit sequence knowledge (Jones and McLaren, 2009, Willingham, 1998). Here too, shorter intervals between successive responses may reduce the contribution of explicit knowledge (Cleeremans and Sarrazin, 2007, Destrebecqz and Cleeremans, 2001). So, in both keying tasks skill is probably based primarily on implicit knowledge while explicit knowledge improves performance to a limited extent.
While, as said, motor chunks do not seem to develop in a typical SRT task, one may wonder whether, vice versa, the associations that are assumed to underlie skill in the SRT task may perhaps develop in the DSP task. This seems reasonable as practicing DSP sequences initially involves responding to key-specific stimuli, just like in the SRT task. Only in regular DSP tasks the development of associations may be overshadowed by the more effective and resource efficient chunking mode.
In retrospect, several DSP studies provide indications that participants may have sometimes produced their sequences in the associative mode. First, when amidst unfamiliar sequences participants suddenly were to execute a familiar sequence, execution rate gradually increased during its execution (Verwey, 2003b, Exp. 1). Analyses of the RT distributions of the moderately fast responses in these familiar sequences showed three peaks. The fast RTs were attributed to the use of motor chunks (thus indicating performance in the chunking mode); slow RTs corresponded to responding to key-specific stimuli (performance in the reaction mode). The – at that time unexplained – intermediate peak suggested a third processing mode. This peak could well indicate the use of the associative mode in which participants still responded to individual stimuli but benefitted from the priming by previous stimuli and responses.
A second indication for the development of an associative mode in the DSP task comes from recent aging studies. Shea, Park, and Wilde Braden (2006) established that elderly over 64 did not seem to use motor chunks in a sequential lever movement task, while young adults did. Two later studies confirmed this idea in the context of the DSP (keying) task: Most elderly over 74, and a substantial proportion of middle-aged participants between 55 and 62 years, seemed not to engage in motor chunking but rather continued executing DSP sequences by responding to key-specific stimuli (Verwey, 2010, Verwey et al., 2011). Importantly, in line with execution in an associative mode the older participants were still faster with their familiar than with unfamiliar sequences (cf. Shea et al., 2006). A particularly interesting observation in the above studies of Verwey and colleagues was that both older groups showed the benefit of the familiar over the unfamiliar sequences to gradually increase with position in the sequences. This is interesting because this finding has been observed with word pronunciation too (Fowler, 1981, Huggins, 1978, Lehiste, 1970). This led to the notion that associative learning allows activation to accumulate across successive movements (MacKay, 1982, Verwey, 1994). Such an increasing benefit cannot be observed in SRT tasks in which behavior at the start of the sequence is normally not analyzed.
Nevertheless, there also is an indication that the associative and chunking modes involve different execution mechanisms that may not co-develop in the context of the DSP task. Verwey (2003b; Exp. 2) showed little transfer from a DSP to an SRT task when familiar DSP sequences were hidden in an SRT task. In contrast, inserting a familiar SRT sequence into another SRT sequence did improve performance (Curran, 1997, Stadler, 1993). This suggests that practicing a DSP task does not produce an associative mode that automatically facilitates execution of that sequence.
The aim of the present study was determining whether the associative mode develops when participants practice a discrete keying sequence which prompts chunk-based control. Reasoning was that such an associative mode always develops in that situation, but is typically overshadowed by the chunking mode; if we can disable the chunking mode the contribution of the associative mode should emerge.
Participants practiced a typical DSP task with two 6-key sequences, and then performed in a test phase with three conditions. First, the pure-familiar condition involved execution of the familiar sequences in the chunking mode, just like in the practice phase. Second, the mixed-familiar condition involved these familiar sequences too, but now most sequences included two stimuli that deviated from the practiced order. We called these stimuli deviants. Participants were informed about these deviants and were instructed to attend to each stimulus before responding. In fact, 75% of the familiar sequences included two randomly determined stimuli across positions 2 to 6 (the first stimulus was never changed), while the remaining 25% of the sequences did not involve deviants. These unchanged sequences were the primary focus of the present study as they were expected not to involve the chunking mode as every deviant would then unavoidably be followed by an error. Third, in the mixed-unfamiliar condition participants performed unfamiliar sequences. Here, too, 75% of the sequences included two random deviants relative to the two (unfamiliar) fixed stimulus orders. As these sequences were unfamiliar they were assumed to be carried out in reaction mode.
The research question in the present study concerned whether the repeated occurrence of deviants in the mixed-familiar condition would cause the unchanged sequences in this condition to be carried out in the associative mode rather than in the reaction mode. Execution in associative mode would be indicated by a higher execution rate of unchanged sequences in the mixed-familiar than in the mixed-unfamiliar condition. Indications for execution of familiar sequences in the associative mode would support the notion that practice in the DSP task induces not only motor chunking but also the associations assumed to be responsible for skill in the SRT task. We assessed also the extent to which explicit knowledge of the keying sequences contributes to the use of motor chunks and the associative mode.
Section snippets
Participants
Twenty-four undergraduate students took part (average age of 21.2, range 18–26 years; 12 women) in exchange for course credits. The study was approved by the ethics committee of Faculty of Behavioral Sciences of the University of Twente.
Task, sequences and stimuli
Six black 9 × 9 mm placeholders were displayed on a computer display with a white background. Between each placeholder there was a 7 mm distance, except between the third and the fourth placeholder where a 22 mm gap was presented to mimic placement of the (DFG and
Results
Sequences with an error (which resulted in immediate sequence abortion) and the first two sequences of each (sub-) block were excluded from the response time (RT) analyses.
Three processing modes
The literature contains several indications that, while in many tasks guidance of sequential motor skills is internal in that movement-specific stimuli are not required (e.g., Goldberg, 1985, Hikosaka et al., 1999), in some other tasks movement sequences are still controlled externally in that individual responses are guided by movement-specific stimuli (e.g., Cohen & Poldrack, 2008). Our review of the research literature on the DSP and the SRT tasks suggests that even the quite similar keying
Acknowledgments
We would like to thank Daniel Bausch, Stefan Becker, Hendrik Müller, and Timo Brandenstein for running the experiment. Elger Abrahamse was supported by the Netherlands Organisation for Scientific Research (NWO) under contract number 446-10-025, and by the Research Foundation — Flanders (FWO) under contract number 12C4712N.
References (63)
- et al.
Time, action, and consciousness
Human Movement Science
(2007) - et al.
Neuroplasticity subserving motor skill learning
Neuron
(2011) - et al.
Sequential congruency effects in implicit sequence learning
Consciousness and Cognition
(2009) - et al.
Chunking by colors: Assessing discrete learning in a continuous serial reaction-time task
Acta Psychologica
(2011) - et al.
Coding of on-line and pre-planned movement sequences
Acta Psychologica
(2010) - et al.
Representation of movement sequences is related to task characteristics
Acta Psychologica
(2009) - et al.
Attentional requirements of learning: Evidence from performance measures
Cognitive Psychology
(1987) - et al.
Structural factors in patterned finger tapping
Acta Psychologica
(1982) - et al.
Learning and production of movement sequences: Behavioral, neurophysiological, and modeling perspectives
Human Movement Science
(2004) - et al.
The latency and duration of rapid movement sequences: Comparisons of speech and typewriting
Evidence for the development of concurrent processing in a sequential keypressing task
Acta Psychologica
Concatenating familiar movement sequences: The versatile cognitive processor
Acta Psychologica
Diminished motor skill development in elderly: Indications for limited motor chunk use
Acta Psychologica
Segmentation of short keying sequences does not spontaneously transfer to other sequences
Human Movement Science
Representing serial action and perception
Psychonomic Bulletin & Review
A closed-loop theory of motor learning
Journal of Motor Behavior
Consciousness: The radical plasticity thesis
Implicit learning and concsciousness: A graded, dynamic perspective
Automaticity in motor sequence learning does not impair response inhibition
Psychonomic Bulletin & Review
Effects of age on implicit sequence learning: Accounting for sequence structure and explicit knowledge
Psychological Research
Attentional and nonattentional forms of sequence learning
Journal of Experimental Psychology: Learning, Memory, and Cognition
Can sequence learning be implicit? New evidence with the process dissociation procedure
Psychonomic Bulletin & Review
Measuring unconscious knowledge: Distinguishing structural knowledge and judgment knowledge
Psychological Research
A relationship between coarticulation and compensatory shortening
Phonetica
Supplementary motor area structure and function: Review and hypotheses
The Behavioral and Brain Sciences
Parallel neural networks for learning sequential procedures
Trends in Neurosciences
Speech timing and intelligibility
The principles of psychology
Taking patterns for chunks: Is there any evidence of chunk learning in continuous serial reaction-time tasks?
Psychological Research
Human sequence learning under incidental and intentional conditions
Journal of Experimental Psychology. Animal Behavior Processes
Movement control in skilled motor performance
Psychological Bulletin
Cited by (47)
Neural mechanisms for learning hierarchical structures of information
2021, Current Opinion in Neurobiology