Cook et al. make the case that sensorimotor learning plays an important role in the origin of mirror neurons (MNs) and that it is therefore essential to pay careful attention to an organism's developmental history and to couple this understanding with rigorous systems-level theory and experimentation. The putative role of MNs in speech perception offers both a cautionary tale of making too much of a surface analogy and a demonstration of the fertility of a systems-level developmental approach to human behavior.
Speech perception is one of the many phenomena claimed to be related to MNs. This link is most commonly made through the Motor Theory (MT) of speech perception, an influential theory that posits that humans perceive speech not as sounds, per se, but as the intended phonetic gestures of the speaker. The central claim is that speech production and perception share common processes and representation, drawn from a linguistic module evolved specifically for communication (Liberman et al. Reference Liberman, Cooper, Shankweiler and Studdert-Kennedy1967; Liberman & Mattingly Reference Liberman and Mattingly1985). One virtue of MT is that it provides an explanation for how listeners bridge the gap between the variable, continuous acoustic speech signal and discrete, relatively invariant linguistic representations. Another virtue is that it explicitly provides a common representation for production and perception – phonetic gestures (or articulations). Empirical tests of MT's predictions have provided mixed support, at best (e.g., Kluender et al. Reference Kluender, Lotto, Holt, Greenberg and Ainsworth2005; Lotto & Holt Reference Lotto and Holt2006; Massaro & Chen Reference Massaro and Chen2008), but the discovery of MNs has led to a rebirth of MT, influencing research in the neuroscience of speech and language processing, speech development, and language evolution (Fogassi & Ferrari Reference Fogassi and Ferrari2007; Galantucci et al. Reference Galantucci, Fowler and Turvey2006).
Mirror neurons have been taken as evidence for MT because they provide neural confirmation of a perception–production link, leading to proposals that “mirror neurons represent the link between sender and receiver that Liberman postulated” (Rizzolatti & Arbib Reference Rizzolatti and Arbib1998, p.189), and that the MN “system mediates … speech perception” (Rizzolatti & Craighero Reference Rizzolatti and Craighero2004, p. 186). Given the importance of common representations for perception and production in a communication system – what Liberman and Mattingly (Reference Liberman and Mattingly1985) called “parity” – the temptation to attribute MNs with a central role in speech is understandable. However, as cautioned against more generally in the target article, the link between MNs and MT has been based largely on analogy and on the similarity of rather coarse descriptions of MNs and MT.
This alluring analogy fails for several reasons. We have previously argued (Lotto et al. Reference Lotto, Hickok and Holt2009) that it is critical to be clear that MT is much more than just the proposal that there is a link between speech perception and production, or that processes of speech perception and production interact in some manner. There is no debate about these tenets. Auditory cortical regions are activated during speech production (e.g., Price et al. Reference Price, Crinion and MacSweeney2011), and motor regions are activated during speech perception (e.g., Wilson et al. Reference Wilson, Saygin, Sereno and Iacoboni2004). However, the nature of the production–perception link and its significance for either production or perception are not well established. In particular, it is not clear how these links or MNs would solve the problems of mapping variable acoustics to linguistic representations, which first motivated MT. Like the authors of the target article, we argue that the analogy between MN and MT is counterproductive in that it directs research endeavors away from systems-level research that might provide explanations for these mappings.
Nonetheless, there are examples of empirical research and computational modeling in speech perception that provide excellent illustrations of the kinds of systems-level approach taking into account developmental history that can move theories away from analogy and toward explanatory power. Imada et al. (Reference Imada, Zhang, Cheour, Taulu, Ahonen and Kuhl2006), for example, used magnetoencephalography to examine passive listening among newborns, 6-month-olds and 12-month-olds. Across all ages, they observed activation localized to the left superior temporal cortex, responsible for auditory analysis. However, whereas regions localized to left inferior frontal cortex (Broca's area) consistent with speech motor analysis were active among 6- and 12-month-olds passively listening to speech, no inferior frontal activation was observed among newborns. This suggests that cortical regions associated with speech perception and production are not intrinsically linked. Rather, links may emerge with experience. Intriguingly, the developmental timeline of this emergence fits with the onset of imitation and canonical babbling, consistent with a sensorimotor association interpretation.
The target article makes the case that the associative hypothesis suggests that correlated sensorimotor experience drives the development of MN through contiguous, but also contingent, association. Speech is an especially rich stimulus for developing such relationships; by and large, the sounds we utter, we also hear and so there are meaningful correlations among auditory, motor, and somatosensory signals with which to learn perception-action relationships. In the fields of speech perception and production, computational models have been constructive in delineating systems-level theories for how such relationships may be learned. The DIVA (Directions Into Velocities of Articulators) model (Guenther & Vladusich Reference Guenther and Vladusich2012) includes an articulatory control system that is trainable by feedback from auditory and somatosensory control systems. Predictions of the model concerning effects of auditory input on speech motor acquisition and control have been extensively tested, and many of the aspects of the model have been related to plausible neural structures. This is exactly the kind of explanatory developmental model called for by Cook et al. in the target article. Whereas the “neurons” in DIVA are active both during perception and production, this is a consequence of the model's architecture and the nature of the feedback. Nothing is gained in the theory by referring to these neurons as MNs. Likewise, a more recent computational-neural model of sensorimotor integration in speech by Hickok et al. (Reference Hickok, Houde and Rong2011) provides a theoretical foundation for understanding perception–production interactions arising from a feedback-feedforward system with no explicit need for neurons genetically specified to respond to action and perception. These types of models provide strong theoretical bases for the development of neurons with “mirror”-like properties as a result of correlations in perceptual-production experience and the demands of communication.
You have
Access
Cook et al. make the case that sensorimotor learning plays an important role in the origin of mirror neurons (MNs) and that it is therefore essential to pay careful attention to an organism's developmental history and to couple this understanding with rigorous systems-level theory and experimentation. The putative role of MNs in speech perception offers both a cautionary tale of making too much of a surface analogy and a demonstration of the fertility of a systems-level developmental approach to human behavior.
Speech perception is one of the many phenomena claimed to be related to MNs. This link is most commonly made through the Motor Theory (MT) of speech perception, an influential theory that posits that humans perceive speech not as sounds, per se, but as the intended phonetic gestures of the speaker. The central claim is that speech production and perception share common processes and representation, drawn from a linguistic module evolved specifically for communication (Liberman et al. Reference Liberman, Cooper, Shankweiler and Studdert-Kennedy1967; Liberman & Mattingly Reference Liberman and Mattingly1985). One virtue of MT is that it provides an explanation for how listeners bridge the gap between the variable, continuous acoustic speech signal and discrete, relatively invariant linguistic representations. Another virtue is that it explicitly provides a common representation for production and perception – phonetic gestures (or articulations). Empirical tests of MT's predictions have provided mixed support, at best (e.g., Kluender et al. Reference Kluender, Lotto, Holt, Greenberg and Ainsworth2005; Lotto & Holt Reference Lotto and Holt2006; Massaro & Chen Reference Massaro and Chen2008), but the discovery of MNs has led to a rebirth of MT, influencing research in the neuroscience of speech and language processing, speech development, and language evolution (Fogassi & Ferrari Reference Fogassi and Ferrari2007; Galantucci et al. Reference Galantucci, Fowler and Turvey2006).
Mirror neurons have been taken as evidence for MT because they provide neural confirmation of a perception–production link, leading to proposals that “mirror neurons represent the link between sender and receiver that Liberman postulated” (Rizzolatti & Arbib Reference Rizzolatti and Arbib1998, p.189), and that the MN “system mediates … speech perception” (Rizzolatti & Craighero Reference Rizzolatti and Craighero2004, p. 186). Given the importance of common representations for perception and production in a communication system – what Liberman and Mattingly (Reference Liberman and Mattingly1985) called “parity” – the temptation to attribute MNs with a central role in speech is understandable. However, as cautioned against more generally in the target article, the link between MNs and MT has been based largely on analogy and on the similarity of rather coarse descriptions of MNs and MT.
This alluring analogy fails for several reasons. We have previously argued (Lotto et al. Reference Lotto, Hickok and Holt2009) that it is critical to be clear that MT is much more than just the proposal that there is a link between speech perception and production, or that processes of speech perception and production interact in some manner. There is no debate about these tenets. Auditory cortical regions are activated during speech production (e.g., Price et al. Reference Price, Crinion and MacSweeney2011), and motor regions are activated during speech perception (e.g., Wilson et al. Reference Wilson, Saygin, Sereno and Iacoboni2004). However, the nature of the production–perception link and its significance for either production or perception are not well established. In particular, it is not clear how these links or MNs would solve the problems of mapping variable acoustics to linguistic representations, which first motivated MT. Like the authors of the target article, we argue that the analogy between MN and MT is counterproductive in that it directs research endeavors away from systems-level research that might provide explanations for these mappings.
Nonetheless, there are examples of empirical research and computational modeling in speech perception that provide excellent illustrations of the kinds of systems-level approach taking into account developmental history that can move theories away from analogy and toward explanatory power. Imada et al. (Reference Imada, Zhang, Cheour, Taulu, Ahonen and Kuhl2006), for example, used magnetoencephalography to examine passive listening among newborns, 6-month-olds and 12-month-olds. Across all ages, they observed activation localized to the left superior temporal cortex, responsible for auditory analysis. However, whereas regions localized to left inferior frontal cortex (Broca's area) consistent with speech motor analysis were active among 6- and 12-month-olds passively listening to speech, no inferior frontal activation was observed among newborns. This suggests that cortical regions associated with speech perception and production are not intrinsically linked. Rather, links may emerge with experience. Intriguingly, the developmental timeline of this emergence fits with the onset of imitation and canonical babbling, consistent with a sensorimotor association interpretation.
The target article makes the case that the associative hypothesis suggests that correlated sensorimotor experience drives the development of MN through contiguous, but also contingent, association. Speech is an especially rich stimulus for developing such relationships; by and large, the sounds we utter, we also hear and so there are meaningful correlations among auditory, motor, and somatosensory signals with which to learn perception-action relationships. In the fields of speech perception and production, computational models have been constructive in delineating systems-level theories for how such relationships may be learned. The DIVA (Directions Into Velocities of Articulators) model (Guenther & Vladusich Reference Guenther and Vladusich2012) includes an articulatory control system that is trainable by feedback from auditory and somatosensory control systems. Predictions of the model concerning effects of auditory input on speech motor acquisition and control have been extensively tested, and many of the aspects of the model have been related to plausible neural structures. This is exactly the kind of explanatory developmental model called for by Cook et al. in the target article. Whereas the “neurons” in DIVA are active both during perception and production, this is a consequence of the model's architecture and the nature of the feedback. Nothing is gained in the theory by referring to these neurons as MNs. Likewise, a more recent computational-neural model of sensorimotor integration in speech by Hickok et al. (Reference Hickok, Houde and Rong2011) provides a theoretical foundation for understanding perception–production interactions arising from a feedback-feedforward system with no explicit need for neurons genetically specified to respond to action and perception. These types of models provide strong theoretical bases for the development of neurons with “mirror”-like properties as a result of correlations in perceptual-production experience and the demands of communication.