How often the battles of psychology have to be fought over again each time with heavier armies and bigger trains…
— William James (Reference James1890, Principles of Psychology, Vol. 2, p. 523, footnote)
Although William James wrote this sentence more than a hundred years ago in his ingenious paragraph on ideomotor action, it could have also have been coined in the discussion about the origin and functional role of mirror neurons (MNs). One important contribution of the associative learning account outlined in the target article by Cook et al. has been to situate the finding of MNs in the historical context of psychological theorizing on the relationship of perception and action. Moreover, associative learning provides a powerful approach to explaining the ontogenesis of MNs based on general learning principles. However, a purely associative account of mirror neurons falls short in explaining a number of important findings regarding the modulation and control of the mirror system. In this comment, we therefore outline an extension of associative learning, namely, ideomotor theory that addresses several of these problems.
While the origins of ideomotor theory can be traced back to the beginning of the nineteenth century, the most prominent proponent of ideomotor theory was William James (1890). In its modern form (Greenwald Reference Greenwald1970), ideomotor theory assumes, just like associative learning, that learning promotes the association of sensory and motor codes. However, ideomotor theory states that in the course of learning, additional ideomotor representations are formed that resemble anticipations of the to-be-produced sensory consequences of an action (see our Fig. 1, a–c, for a model of how ideomotor representations are formed).
Figure 1.
Acquisition of an ideomotor representation (adapted from Greenwald Reference Greenwald1970). (a) A stimulus (S) triggers a specific response (R) that leads to a sensory effect (E). (b) After learning, the stimulus will activate an anticipation (e) of the effect that precedes the response. (c) This anticipation (e) becomes conditioned to the response and allows for control of the response. (d) Priming by action observation: a stimulus that resembles the effect of the action (Se) primes the ideomotor representation (e) which activates the response.
According to ideomotor theory, these representations primarily serve a motor control function. We control our actions by anticipating their sensory consequences. Moreover, ideomotor theory predicts a specific form of sensorimotor compatibility, namely, ideomotor compatibility. A stimulus that resembles the anticipation of a sensory action-effect activates the corresponding ideomotor representation (see Fig. 1d). For example, the image of another person opening their hand strongly overlaps with the representation that is used to control the hand-opening movement. Consequently, ideomotor-compatible stimuli can to some degree bypass response selection by directly activating motor programs (Brass et al. Reference Brass, Bekkering and Prinz2001).
As ideomotor representations are conceived as neither uniquely sensory nor motor, they should be localized in dedicated motor control structures that are distinct from primary sensory or motor areas. Such representations can be activated without necessarily leading to overt behavior and are thus likely used for motor planning and prediction (see also the commentary by Keysers, Perrett, and Gazzola for the idea of mirror neurons being involved in predictive coding). This property of ideomotor theory is consistent with human brain imaging studies showing an overlap of brain areas involved in action planning, movement observation, and motor imagery (Grezes & Decety Reference Grèzes and Decety2001).
Another important consequence of ideomotor compatibility is that it can lead to self–other confusion. Because ideomotor-compatible stimuli directly activate representations that are used for motor control, confusion can arise between externally and intentionally triggered motor representations. Accordingly, controlling imitative behavior has been related to brain areas that are involved in the sense of agency and self–other distinction, and dissociated from brain areas involved in controlling interference from overlearned stimulus–response associations (Brass et al. Reference Brass, Derrfuss, Matthes-von Cramon and von Cramon2003; Reference Brass, Derrfuss and von Cramon2005).
One crucial difference between ideomotor representations and simple stimulus–response associations relates to the underlying learning mechanisms. Ideomotor representations evolve from learning the relationship between responses and subsequent sensory effects (R-E learning). In contrast, classical associative learning theories, although concerned with action-outcome contingencies, primarily focus on learning the relationship of responses to those stimuli that precede them (S-R learning). Importantly, most experiments demonstrating that imitative response tendencies can be easily reversed use S-R learning paradigms rather than R-E learning paradigms (e.g., Catmur et al. Reference Catmur, Walsh and Heyes2007). From an ideomotor theory perspective, these findings may reflect that rapid learning strengthens the corresponding S-R associations to such a degree that they temporarily overrule existing ideomotor representations, leading to an advantage of ideomotor-incompatible over compatible mappings (Catmur et al. Reference Catmur, Walsh and Heyes2007).
Another difference between associative learning and ideomotor theory lies in their capacity to deal with specific forms of contextual modulation. Cook et al. outline how associative learning can explain the influence of contextual information on MN responses. In human studies, however, it has been demonstrated that the response of the mirror system is not only sensitive to contextual cues but also to high-level beliefs about the intentionality of the observed action (Liepelt et al. Reference Liepelt, von Cramon and Brass2008b). Ideomotor compatibility can account for such findings, as it is based on the representational overlap between the observed event and the ideomotor representation. Therefore, stimuli that are not perceived as resulting from an intentional action will activate the ideomotor representation to a smaller degree. To our knowledge, a similarly convincing interpretation of such effects from an associative learning perspective is still lacking.
Directly testing for dissociations between associative learning and ideomotor theory proves to be very difficult, as ideomotor theory and associative learning share a common learning phase. In order to demonstrate that ideomotor theory differs from associative learning, one must reveal the anticipatory nature of ideomotor representations. Most tests of ideomotor theory, however, merely show that perceiving learned action effects activate a corresponding motor representation in the observer. This prediction is shared by both approaches. A notable exception to this is a paradigm developed by Kunde (Reference Kunde2001). He showed that when actions are consistently followed by incompatible effects (pressing a right key that is followed by a left stimulus), participants react slower than when actions are consistently followed by compatible effects (pressing a right key that is followed by a right stimulus). In contrast to classical S-R compatibility phenomena, this effect unequivocally originates from the conflict between the previously acquired anticipatory representation of the action and the anticipation of the actual sensory consequence.
To conclude, the associative learning account by Cook and colleagues certainly provides a powerful account of the functions and origins of the mirror system. However, we propose that ideomotor theory provides an important extension of associative learning that is necessary to account for a number of phenomena that are difficult to explain from a purely associative perspective.
How often the battles of psychology have to be fought over again each time with heavier armies and bigger trains…
— William James (Reference James1890, Principles of Psychology, Vol. 2, p. 523, footnote)Although William James wrote this sentence more than a hundred years ago in his ingenious paragraph on ideomotor action, it could have also have been coined in the discussion about the origin and functional role of mirror neurons (MNs). One important contribution of the associative learning account outlined in the target article by Cook et al. has been to situate the finding of MNs in the historical context of psychological theorizing on the relationship of perception and action. Moreover, associative learning provides a powerful approach to explaining the ontogenesis of MNs based on general learning principles. However, a purely associative account of mirror neurons falls short in explaining a number of important findings regarding the modulation and control of the mirror system. In this comment, we therefore outline an extension of associative learning, namely, ideomotor theory that addresses several of these problems.
While the origins of ideomotor theory can be traced back to the beginning of the nineteenth century, the most prominent proponent of ideomotor theory was William James (1890). In its modern form (Greenwald Reference Greenwald1970), ideomotor theory assumes, just like associative learning, that learning promotes the association of sensory and motor codes. However, ideomotor theory states that in the course of learning, additional ideomotor representations are formed that resemble anticipations of the to-be-produced sensory consequences of an action (see our Fig. 1, a–c, for a model of how ideomotor representations are formed).
Acquisition of an ideomotor representation (adapted from Greenwald Reference Greenwald1970). (a) A stimulus (S) triggers a specific response (R) that leads to a sensory effect (E). (b) After learning, the stimulus will activate an anticipation (e) of the effect that precedes the response. (c) This anticipation (e) becomes conditioned to the response and allows for control of the response. (d) Priming by action observation: a stimulus that resembles the effect of the action (Se) primes the ideomotor representation (e) which activates the response.
According to ideomotor theory, these representations primarily serve a motor control function. We control our actions by anticipating their sensory consequences. Moreover, ideomotor theory predicts a specific form of sensorimotor compatibility, namely, ideomotor compatibility. A stimulus that resembles the anticipation of a sensory action-effect activates the corresponding ideomotor representation (see Fig. 1d). For example, the image of another person opening their hand strongly overlaps with the representation that is used to control the hand-opening movement. Consequently, ideomotor-compatible stimuli can to some degree bypass response selection by directly activating motor programs (Brass et al. Reference Brass, Bekkering and Prinz2001).
As ideomotor representations are conceived as neither uniquely sensory nor motor, they should be localized in dedicated motor control structures that are distinct from primary sensory or motor areas. Such representations can be activated without necessarily leading to overt behavior and are thus likely used for motor planning and prediction (see also the commentary by Keysers, Perrett, and Gazzola for the idea of mirror neurons being involved in predictive coding). This property of ideomotor theory is consistent with human brain imaging studies showing an overlap of brain areas involved in action planning, movement observation, and motor imagery (Grezes & Decety Reference Grèzes and Decety2001).
Another important consequence of ideomotor compatibility is that it can lead to self–other confusion. Because ideomotor-compatible stimuli directly activate representations that are used for motor control, confusion can arise between externally and intentionally triggered motor representations. Accordingly, controlling imitative behavior has been related to brain areas that are involved in the sense of agency and self–other distinction, and dissociated from brain areas involved in controlling interference from overlearned stimulus–response associations (Brass et al. Reference Brass, Derrfuss, Matthes-von Cramon and von Cramon2003; Reference Brass, Derrfuss and von Cramon2005).
One crucial difference between ideomotor representations and simple stimulus–response associations relates to the underlying learning mechanisms. Ideomotor representations evolve from learning the relationship between responses and subsequent sensory effects (R-E learning). In contrast, classical associative learning theories, although concerned with action-outcome contingencies, primarily focus on learning the relationship of responses to those stimuli that precede them (S-R learning). Importantly, most experiments demonstrating that imitative response tendencies can be easily reversed use S-R learning paradigms rather than R-E learning paradigms (e.g., Catmur et al. Reference Catmur, Walsh and Heyes2007). From an ideomotor theory perspective, these findings may reflect that rapid learning strengthens the corresponding S-R associations to such a degree that they temporarily overrule existing ideomotor representations, leading to an advantage of ideomotor-incompatible over compatible mappings (Catmur et al. Reference Catmur, Walsh and Heyes2007).
Another difference between associative learning and ideomotor theory lies in their capacity to deal with specific forms of contextual modulation. Cook et al. outline how associative learning can explain the influence of contextual information on MN responses. In human studies, however, it has been demonstrated that the response of the mirror system is not only sensitive to contextual cues but also to high-level beliefs about the intentionality of the observed action (Liepelt et al. Reference Liepelt, von Cramon and Brass2008b). Ideomotor compatibility can account for such findings, as it is based on the representational overlap between the observed event and the ideomotor representation. Therefore, stimuli that are not perceived as resulting from an intentional action will activate the ideomotor representation to a smaller degree. To our knowledge, a similarly convincing interpretation of such effects from an associative learning perspective is still lacking.
Directly testing for dissociations between associative learning and ideomotor theory proves to be very difficult, as ideomotor theory and associative learning share a common learning phase. In order to demonstrate that ideomotor theory differs from associative learning, one must reveal the anticipatory nature of ideomotor representations. Most tests of ideomotor theory, however, merely show that perceiving learned action effects activate a corresponding motor representation in the observer. This prediction is shared by both approaches. A notable exception to this is a paradigm developed by Kunde (Reference Kunde2001). He showed that when actions are consistently followed by incompatible effects (pressing a right key that is followed by a left stimulus), participants react slower than when actions are consistently followed by compatible effects (pressing a right key that is followed by a right stimulus). In contrast to classical S-R compatibility phenomena, this effect unequivocally originates from the conflict between the previously acquired anticipatory representation of the action and the anticipation of the actual sensory consequence.
To conclude, the associative learning account by Cook and colleagues certainly provides a powerful account of the functions and origins of the mirror system. However, we propose that ideomotor theory provides an important extension of associative learning that is necessary to account for a number of phenomena that are difficult to explain from a purely associative perspective.