Both rule- and instance-based learning rely on information-processing models of motor control in which a memory representation of the motor behavior drives the execution, or production, of that motor act. These motor control models rely on a hierarchy in which upper levels control lower levels by organizing and fine-tuning the representation as processing travels down the levels (Hulstijn & Van Galen, 1983; D. E. Meyer & Gordon, 1985; Rogers & Storkel, 1998; Rosenbaum, 1990; Schmidt & Wrisberg, 2004; Schmidt, 1988). Generally, information- processing models of motor control have four levels: cognitive input, response selection, response programming, and execution (Schmidt & Wrisberg, 2004; Schmidt, 1988).
During the cognitive input stage, stimulus properties are identified to determine whether the stimulus is auditory, visual, kinesthetic, proprioceptive, or some combination of these properties (Miller & Ulrich, 1998; Schmidt & Wrisberg, 2004). From a communication standpoint, the input signal may derive from a communication partner (e.g., a spoken auditory signal) or an individual’s own intent to communicate. Factors influencing the cognitive input stage include: attention (Ono, 1990; Rhodes, Bullock, Verwey, Averbeck, & Page, 2004), working memory (Rhodes et al., 2004), and stimulus characteristics (e.g., clarity, complexity, intensity; Schmidt, 1988). Most speech production models would also place linguistic processing at this level (e.g., Van der Merwe, 1997). Once the stimulus input is received, a decision is made to act on the information, and a motor response is selected.
Response selection involves selecting the memory representation that will direct the intention to act put forth by the cognitive input stage (Curtis, Rao, & D’Esposito, 2004). Verwey
(2001) considered this memory representation a “goal structure of action” (p. 71). The overall goal of the movement is defined during this stage, as well as the general movement parameters for achieving the goal (Curtis et al., 2004). This memory representation is typically depicted as a motor program4 (Klapp, 1995; McCann & Johnston, 1992); however, an instance-based interpretation is also feasible if the memory representation is conceptualized as a knowledge base of instances. These potential response alternatives are depicted in Figure 7 in which the responses selected can be viewed as a collection of individual exemplars or a schema.
4 The term “motor program” as used here is more general in its scope than as used in Chapter 2 (e.g.,
GMP). The term here refers to a general plan of action for achieving a goal state. More finite “motor programs” are established in the response programming stage, in which the sequence of movements and other factors contribute to a specific movement pattern.
Figure 7: Potential response alternatives
Note: Adapted from Journal of Experimental Psychology: General, 109(4), Rosenbaum, Human movement initiation: Specification of arm, direction, and extent, p. 447, 1980, with permission from American Psychology Association.
Response programming organizes and specifies the parameters of the selected representation (Klapp, 1995). Programming of the selected representation may involve specifying the spatial and temporal parameters of the movement, as well as stringing together several representations, or chunks, into larger, more complex movement patterns (Hulstijn & Van Galen, 1983; Klapp, 1995, 1996; Schmidt & Wrisberg, 2004). Evidence of dissociation between programming specification and serial ordering has been reported within the response programming stage (Deger & Ziegler, 2002; Klapp, 1995; Maas, Robin, Wright, & Ballard, 2008). The majority of reaction time studies in the speech production literature are focused on response programming (e.g., Deger & Ziegler, 2002; Maas & Mailend, 2012; Maas, Robin, Wright, et al., 2008; D. E. Meyer & Gordon, 1985; Rogers & Storkel, 1998).
Response Alternatives Exemplar Knowledge Base A=1, B=1, C=1 A=1, B=1, C=2 A=1, B=1, C=3 Motor Schema/Program A=1, B=1, C = ?
The content of the response selection and programming stages changes over the course of learning. Early in practice (or when performing novel movements) specific temporal and spatial targets, as well as motor commands to specific muscle groups, are not yet coded. These specifications are determined during the programming stage of motor performance (Klapp, 1995). With practice, chunking occurs and the specifications of the movement parameters become associated with the representation itself (Schmidt & Wrisberg, 2004). Thus, later in practice (or when performing a well-learned task) chunks, or pre-loaded motor representations, may be selected and sent on for serial ordering only in the programming stage (Hulstijn & Van Galen, 1983; Maas & Mailend, 2012; Verwey, 2001).
At the lowest level of the model is motor execution, in which muscles contract and the behavior is realized physiologically (Hulstijn & Van Galen, 1983; Schmidt & Wrisberg, 2004). Motor output can be described by how well the movement performance met the desired goal (i.e., accuracy or error values), the kinematics of the movement (e.g., force, velocity, acceleration), or the muscular activity during the movement (e.g., electromyography; Schmidt & Lee, 2005). Analysis of an acoustic signal is an important variable of motor execution in speech production, as the movement goal results in an acoustic signal a listener, or communication partner, can understand (Guenther et al., 1998; Perkell et al., 1995).
Different models of motor control employ different terminology to describe these four levels, and may only incorporate certain levels depending on the specificity of the model. However, generally the four main components of motor performance are assumed within each model. Additionally, it is debated whether the transmission of information between stages within each model is serial (e.g., Henry & Rogers, 1960; Hulstijn & Van Galen, 1983; D. E. Meyer & Gordon, 1985) or parallel (e.g., Miller & Ulrich, 1998; Verwey, 1995, 1999, 2001). For ease of
explanation, I have presented the information-processing model of motor control as a serial model, in which each stage of the model processes information in a step-wise fashion. In this way, only one level of the model is operating at a single moment in time. However, there is a variety of research suggesting parallel processing can occur between stages, especially between the levels of response selection and programming (e.g., Rhodes et al., 2004; Verwey, 1995, 1999, 2001). For example, while end-sequence movements are finalized in the response programming buffer (i.e., ready for execution), upcoming motor responses are being selected (Verwey, 1999, 2001). The purpose of presenting these models is not to debate whether models of motor control have serial or parallel processing. The purpose, instead, is to outline a general approach to models of motor control and how their components are assessed (see Section 4.2.2) to inform an evaluation of rule- versus instance-based learning in the motor literature.
Appendix A depicts two information-processing motor control models. Limb motor control is illustrated in Figure 30, in which the “motor program” is situated at the general level of execution. Input from the higher levels of response selection and programming develop the motor program. Thus, in this context, the motor program encapsulates the memory representation, the kinematic parameters, muscle specification, and serial order of the movement (Schmidt & Wrisberg, 2004). Van der Merwe’s (1997) model of speech production is depicted in Figure 31. The highest level within the model is composed of cognitive-linguistic input, followed by a large motor “planning” stage, then programming, and finally an execution stage for speech output. The motor planning stage portrays several selection processes, including phonemic and phonetic selections, which are sequenced together. The “programming” stage of this model specifies the muscles required to complete the phonetic commands. The parsing of speech and language in this model distinguishes cognitive/linguistic planning (language) and motor
planning/programming (speech). This division in the planning stage is not noted in other speech production models in which the speech components are maintained in the programming or execution stages (e.g., Levelt et al., 1999).