Abstraction and selective tuning for conspecifics (lack thereof)
In both correlational and causal studies, human MNS has been linked to the processing of point-light biological motion stimuli, which depict actions through motion cues (Gilaie-Dotan et al. Reference Gilaie-Dotan, Kanai, Bahrami, Rees and Saygin2013; Saygin et al. Reference Saygin, Wilson, Hagler, Bates and Sereno2004; van Kemenade et al. Reference van Kemenade, Muggleton, Walsh and Saygin2012). In contrast, macaque MNs do not even respond to videos of actions (Ferrari et al. Reference Ferrari, Gallese, Rizzolatti and Fogassi2003), except possibly after extensive training (Caggiano et al. Reference Caggiano, Fogassi, Rizzolatti, Pomper, Thier, Giese and Casile2011). Although differences between humans and nonhuman primates could also have cross-methodology sources (e.g., firing rates, vs. LFPs, vs. BOLD fMRI), contrasting the ease with which human observers process point-light stimuli, with the daunting undertaking of getting nonhuman primates to perform even the simplest tasks with them (Vangeneugden et al. Reference Vangeneugden, Vancleef, Jaeggli, VanGool and Vogels2009), it appears there may be cross-species differences. The ability of the human MNS to extrapolate to such abstract representations is difficult to explain with a specific genetic adaptation, but fits rather well as a natural by-product of the human ability to effectively use abstractions in associative learning (cf. language, see next), itself an emergent outcome of domain-general aspects of primate evolution.
Consider also whether the MNS is adapted specifically for conspecifics. Although one study has claimed the human MNS is “mirror only for other humans,” further studies have reported that human MNS also responds to robot actions (e.g., Cross et al. Reference Cross, Liepelt, Hamilton, Parkinson, Ramsey, Stadler and Prinz2012; Gazzola et al. Reference Gazzola, Rizzolatti, Wicker and Keysers2007; Oberman et al. Reference Oberman, McCleery, Ramachandran and Pineda2007), as do subdivisions of F5 in the macaque (Nelissen et al. Reference Nelissen, Luppino, Vanduffel, Rizzolatti and Orban2005). EEG and fMRI studies involving highly human-like androids indicate that the human MNS is not straightforwardly tuned to human motion or human appearance (Saygin et al. Reference Saygin, Chaminade, Ishiguro, Driver and Frith2012; Urgen et al. Reference Urgen, Plank, Ishiguro, Poizner and Saygin2013). Overall, in terms of response properties, it appears that MNS is not highly selective to conspecifics, or even living things. Such data, and more generally, the oft-noted insensitivity of MNs to perceptual factors, are much more consistent with an adaptationist account than a genetic account of MNs. For why should the brain have a specially evolved system for understanding the actions of others that is also responsive to stimuli that did not exist in the natural environment until very recently, such as twentieth century point-light displays and twenty-first century Japanese androids?
Notes from speech and language
Language has been one of the most paradigmatic areas in which nature versus nurture and domain-specific versus domain-general debates took place (Elman et al. Reference Elman, Bates, Johnson, Karmiloff-Smith, Parisi and Plunkett1996). We have long held that language is best viewed as a complex skill that emerges from adaptations of domain-general sensorimotor neural systems (Dick et al. Reference Dick, Saygin, Moineau, Aydelott and Bates2004; Reference Dick, Saygin, Galati, Pitzalis, Bentrovato, D'Amico, Wilson, Bates and Pizzamiglio2007; Saygin et al. Reference Saygin, Dick, Wilson, Dronkers and Bates2003), nicely summarized by the late Elizabeth Bates as “language is a new machine built out of old parts” (Bates et al. Reference Bates, Bretherton and Snyder1988).
The field of speech and language has lived through decades of the “poverty of the stimulus” argument regarding the purported necessity for innately specified, special mechanisms and representations (see commentary by Holt & Lotto, this BBS issue). But if anything, the emerging consensus is that language has parasitized a remarkably plastic yet determinedly primate brain that is organized along sensorimotor and not domain-specific lines. We suggest that the continuing saga of mirror neurons may follow a similar dramatic arc.
The domain of speech provides a compelling example of how MNs might develop in a simple “fire together wire together” learning mechanism with simple (yet realistic) anatomical constraints. In a computational model of infant babbling, Westermann and Miranda (Reference Westermann and Miranda2002; Reference Westermann and Miranda2004) showed that a network that produces babble (permutations of muscle movement combinations), and simultaneously “hears” its acoustical output, will spontaneously form MNs for speech. Once the computational baby has babbled sufficiently, when it then hears a vowel, the auditory representation map of this vowel will activate the motor representation that has been linked up to it via Hebbian learning.
A strong prediction of this early associative model of the emergence of MNs for speech was found in an fMRI study: Enhanced activation for passively heard syllables was observed in a premotor region that was strongly activated during production of the same syllables (Wilson et al. Reference Wilson, Saygin, Sereno and Iacoboni2004). Although this finding was taken by some as confirmation of the motor theory of speech perception, subsequent studies have shown that mirror-like responses can emerge in motor regions as a product of more general audiomotor expertise, as would be predicted by associative accounts. When listening to dramatic speech and solo violin playing, professional actors show considerably more activation for speech in premotor areas, including that reported in Wilson et al. (Reference Wilson, Saygin, Sereno and Iacoboni2004). Conversely, violinists listening to the same stimuli show the opposite pattern – much greater activation for listening to violin excerpts than for speech (Dick et al. Reference Dick, Lee, Nusbaum and Price2011).
Overall, mirror-like responses, rather than being specific genetic adaptations, are more likely to be evidence for the remarkable functional plasticity that allows primates – especially humans – to learn and master complex contingencies via domain-general mechanisms.
You have
Access
The discovery of mirror neurons (MNs) proffered a tempting solution for the difficult theoretical and neurocomputational problem of linking self and other (Barresi & Moore Reference Barresi and Moore1996). MNs were regarded in some quarters as a neuroscientific silver bullet, one providing the evolutionary and neural basis of social cognition, empathy, theory of mind, language, and civilization. After this period of MN-Mania, inevitably there has been a fierce counter-MN-reformation, with some acolytes vociferously denying the mere existence of MNs. By pleasant contrast to either extreme, the target article offers a sober and reasoned framework for thinking about the origins and emergence of the Mirror Neuron System (MNS). Here, we offer additional evidence that may be explained with an associative MN account, but are problematic to account for via specific innate mechanisms. First, the human MNS seems capable of abstraction and generalization, as evidenced by its involvement in processing novel visual stimuli that have not existed through evolution and are thus hard to square with built-in, genetically coded MNs. Second, a straightforward associative model of early speech production accurately predicts the existence of MN-like responses for speech; and domain-specific mirror-like responses emerge in response to intensive training in producing and perceiving instrumental sounds.
Abstraction and selective tuning for conspecifics (lack thereof)
In both correlational and causal studies, human MNS has been linked to the processing of point-light biological motion stimuli, which depict actions through motion cues (Gilaie-Dotan et al. Reference Gilaie-Dotan, Kanai, Bahrami, Rees and Saygin2013; Saygin et al. Reference Saygin, Wilson, Hagler, Bates and Sereno2004; van Kemenade et al. Reference van Kemenade, Muggleton, Walsh and Saygin2012). In contrast, macaque MNs do not even respond to videos of actions (Ferrari et al. Reference Ferrari, Gallese, Rizzolatti and Fogassi2003), except possibly after extensive training (Caggiano et al. Reference Caggiano, Fogassi, Rizzolatti, Pomper, Thier, Giese and Casile2011). Although differences between humans and nonhuman primates could also have cross-methodology sources (e.g., firing rates, vs. LFPs, vs. BOLD fMRI), contrasting the ease with which human observers process point-light stimuli, with the daunting undertaking of getting nonhuman primates to perform even the simplest tasks with them (Vangeneugden et al. Reference Vangeneugden, Vancleef, Jaeggli, VanGool and Vogels2009), it appears there may be cross-species differences. The ability of the human MNS to extrapolate to such abstract representations is difficult to explain with a specific genetic adaptation, but fits rather well as a natural by-product of the human ability to effectively use abstractions in associative learning (cf. language, see next), itself an emergent outcome of domain-general aspects of primate evolution.
Consider also whether the MNS is adapted specifically for conspecifics. Although one study has claimed the human MNS is “mirror only for other humans,” further studies have reported that human MNS also responds to robot actions (e.g., Cross et al. Reference Cross, Liepelt, Hamilton, Parkinson, Ramsey, Stadler and Prinz2012; Gazzola et al. Reference Gazzola, Rizzolatti, Wicker and Keysers2007; Oberman et al. Reference Oberman, McCleery, Ramachandran and Pineda2007), as do subdivisions of F5 in the macaque (Nelissen et al. Reference Nelissen, Luppino, Vanduffel, Rizzolatti and Orban2005). EEG and fMRI studies involving highly human-like androids indicate that the human MNS is not straightforwardly tuned to human motion or human appearance (Saygin et al. Reference Saygin, Chaminade, Ishiguro, Driver and Frith2012; Urgen et al. Reference Urgen, Plank, Ishiguro, Poizner and Saygin2013). Overall, in terms of response properties, it appears that MNS is not highly selective to conspecifics, or even living things. Such data, and more generally, the oft-noted insensitivity of MNs to perceptual factors, are much more consistent with an adaptationist account than a genetic account of MNs. For why should the brain have a specially evolved system for understanding the actions of others that is also responsive to stimuli that did not exist in the natural environment until very recently, such as twentieth century point-light displays and twenty-first century Japanese androids?
Notes from speech and language
Language has been one of the most paradigmatic areas in which nature versus nurture and domain-specific versus domain-general debates took place (Elman et al. Reference Elman, Bates, Johnson, Karmiloff-Smith, Parisi and Plunkett1996). We have long held that language is best viewed as a complex skill that emerges from adaptations of domain-general sensorimotor neural systems (Dick et al. Reference Dick, Saygin, Moineau, Aydelott and Bates2004; Reference Dick, Saygin, Galati, Pitzalis, Bentrovato, D'Amico, Wilson, Bates and Pizzamiglio2007; Saygin et al. Reference Saygin, Dick, Wilson, Dronkers and Bates2003), nicely summarized by the late Elizabeth Bates as “language is a new machine built out of old parts” (Bates et al. Reference Bates, Bretherton and Snyder1988).
The field of speech and language has lived through decades of the “poverty of the stimulus” argument regarding the purported necessity for innately specified, special mechanisms and representations (see commentary by Holt & Lotto, this BBS issue). But if anything, the emerging consensus is that language has parasitized a remarkably plastic yet determinedly primate brain that is organized along sensorimotor and not domain-specific lines. We suggest that the continuing saga of mirror neurons may follow a similar dramatic arc.
The domain of speech provides a compelling example of how MNs might develop in a simple “fire together wire together” learning mechanism with simple (yet realistic) anatomical constraints. In a computational model of infant babbling, Westermann and Miranda (Reference Westermann and Miranda2002; Reference Westermann and Miranda2004) showed that a network that produces babble (permutations of muscle movement combinations), and simultaneously “hears” its acoustical output, will spontaneously form MNs for speech. Once the computational baby has babbled sufficiently, when it then hears a vowel, the auditory representation map of this vowel will activate the motor representation that has been linked up to it via Hebbian learning.
A strong prediction of this early associative model of the emergence of MNs for speech was found in an fMRI study: Enhanced activation for passively heard syllables was observed in a premotor region that was strongly activated during production of the same syllables (Wilson et al. Reference Wilson, Saygin, Sereno and Iacoboni2004). Although this finding was taken by some as confirmation of the motor theory of speech perception, subsequent studies have shown that mirror-like responses can emerge in motor regions as a product of more general audiomotor expertise, as would be predicted by associative accounts. When listening to dramatic speech and solo violin playing, professional actors show considerably more activation for speech in premotor areas, including that reported in Wilson et al. (Reference Wilson, Saygin, Sereno and Iacoboni2004). Conversely, violinists listening to the same stimuli show the opposite pattern – much greater activation for listening to violin excerpts than for speech (Dick et al. Reference Dick, Lee, Nusbaum and Price2011).
Overall, mirror-like responses, rather than being specific genetic adaptations, are more likely to be evidence for the remarkable functional plasticity that allows primates – especially humans – to learn and master complex contingencies via domain-general mechanisms.