One challenge encountered when reverse engineering the brain is that the same function (e.g., categorizing an enjoyment smile) can be carried out by vastly different mechanisms (e.g., visual feature analysis, conceptual processing, or embodied simulation). This gives the brain tremendous flexibility and the ability to compensate for failed mechanisms. To the detriment of the scientist, it also makes it difficult to experimentally isolate the mechanism(s) underlying a given function (e.g., the categorization of smiles) and identify the conditions under which a given function will be carried out by one mechanism versus another. With this in mind, Niedenthal et al. should be commended for their trailblazing contribution that specifies both the conditions under which embodied simulation is likely to occur (e.g., following eye contact) and the neural-physiological components of this intricate and counterintuitive mechanism.
When appreciating such an advancement, it is worth keeping in mind that the primary explanatory value of embodied approaches to knowledge representation (e.g., Barsalou Reference Barsalou1999; Glenberg Reference Glenberg1997) rested in their ability to address the symbol grounding problem (Harnad Reference Harnad1990) faced by “propositional” (aka “amodal”) accounts of mental representation (e.g., Landauer & Dumais Reference Landauer and Dumais1997), in which meaning is represented by arbitrary symbols (e.g., binary digits) that do not retain any of the properties of the sensorimotor states that gave rise to them (see review in Markman & Dietrich Reference Markman and Dietrich2000). Similar to “analogical” and “modal” accounts, embodied models propose that meaning is constituted by (or “grounded in”) sensorimotor information, such as that furnished by facial expressions. Since their advent, embodied models have been challenged on empirical grounds. Regarding emotion, for example, an extensive study revealed that people with bilateral facial paralysis stemming from Moebius syndrome can categorize the facial expressions of others without difficulty (Bogart & Matsumoto Reference Bogart and Matsumoto2010). Moreover, the ability of subjects to express facial expressions does not influence accuracy. Similarly, in semantics, it remains unclear whether lesions of action-related regions of the brain lead to impairments regarding the semantic-conceptual knowledge of action: Patients suffering from deficits due to brain lesions can show impairments in the way they use certain objects but, revealing the abilities of a “disembodied” conceptual system, can nonetheless name those objects, or even recognize the pantomimes associated with them (Mahon & Caramazza Reference Mahon and Caramazza2008).
In response to such challenges, embodied approaches now propose that sensorimotor information plays a more humble role: In the target article, Niedenthal et al. portray sensorimotor information not as that which constitutes meaning, but merely as auxiliary information that is “referred to” when semantic processing is challenged, as when categorizations are difficult to make. In these accounts, sensorimotor information is demoted to a kind of supplementary information that, just like any kind of information, can at times facilitate or interfere with semantic processing. When the strategy of embodied simulation fails, it is the conceptual system that arrives at the scene to solve the categorization, much like a deus ex machina.
The question now is whether the conceptual system is always necessary for categorization. From one standpoint, the answer is yes: Until a token facial expression is typed as a member of a given category (an inherently conceptual process), the organism cannot know (consciously or unconsciously) the meaning of the observed expression. If the conceptual system must always be at the scene, then what is its nature and how does it solve the symbol grounding problem? Embodied approaches were expected to answer this by demonstrating, for example, that disrupting the sensorimotor states in which a conceptual process is grounded will not just modulate the process, but also eradicate it. It seems that, apart from modulations that could be construed as priming or its opposite (i.e., interference), there is no evidence of conceptual processing requiring sensorimotor states. Apart from the aforementioned neurological evidence, this is evident with respect to abstract concepts (e.g., that of if and justice) and everyday phenomena such as imageless thought (e.g., a state of uncertainty or readiness; cf. Hilgard Reference Hilgard1987). Regarding imageless thought, Woodworth (Reference Woodworth1915) concluded, “I know of no reason in neurology or psychology for supposing that the elements of conscious content are contributed solely by the sensory receiving centers” (p. 137).
A way forward? Identifying the limitations of “disembodied” processing
Perhaps a critical next step in theory-building for embodied approaches is to identify the unique contribution(s) of embodied simulation. What is it that a disembodied conceptual system is incapable of doing with respect to semantic processing or the categorization of smiles? Emotion research has broached this topic by revealing some limits of language, a form of propositional processing. Evidence suggests that linguistic processing does not lead to the same kinds of emotional effects and forms of affective learning as modal stimuli (LeDoux Reference LeDoux1996). Olsson and Phelps (Reference Olsson and Phelps2004) demonstrated that, through vicarious classical conditioning, subjects can acquire a learned fear response toward a subliminal stimulus by perceiving someone else receive a shock after being presented with that stimulus. However, simply telling subjects about the contingency between stimuli did not lead to this form of vicarious conditioning to subliminal stimuli. This suggests that the phylogenetically old systems mediating fear conditioning may not understand language, though they can process the meaning of basic perceptual events: “classical conditioning and observational learning … might be supported by an evolutionarily old system that predates the emergence of language” (Olsson & Phelps Reference Olsson and Phelps2007, p. 1099).
That language fails to yield the vicarious conditioning afforded by observational learning may be because language comprehension yields “faint” activations of analogical, sensorimotor-like symbols (Boroditsky & Ramscar Reference Boroditsky and Ramscar2002; Zwaan Reference Zwaan, DeVega, Glenberg and Graesser2008) or, at odds with embodied accounts, because linguistic-semantic knowledge is simply distinct from sensorimotor knowledge. Regarding the pioneering investigation by Niedenthal et al., the next advancement may be to isolate that which only embodied simulations can achieve. It may be that aspects of social mimicry and empathy require embodied simulation, or that the three smile-categorization mechanisms operate differently with respect to subliminal or supraliminal stimuli, with the conceptual deus ex machina being capable of arriving at the scene only when the theater lights of consciousness are on.
One challenge encountered when reverse engineering the brain is that the same function (e.g., categorizing an enjoyment smile) can be carried out by vastly different mechanisms (e.g., visual feature analysis, conceptual processing, or embodied simulation). This gives the brain tremendous flexibility and the ability to compensate for failed mechanisms. To the detriment of the scientist, it also makes it difficult to experimentally isolate the mechanism(s) underlying a given function (e.g., the categorization of smiles) and identify the conditions under which a given function will be carried out by one mechanism versus another. With this in mind, Niedenthal et al. should be commended for their trailblazing contribution that specifies both the conditions under which embodied simulation is likely to occur (e.g., following eye contact) and the neural-physiological components of this intricate and counterintuitive mechanism.
When appreciating such an advancement, it is worth keeping in mind that the primary explanatory value of embodied approaches to knowledge representation (e.g., Barsalou Reference Barsalou1999; Glenberg Reference Glenberg1997) rested in their ability to address the symbol grounding problem (Harnad Reference Harnad1990) faced by “propositional” (aka “amodal”) accounts of mental representation (e.g., Landauer & Dumais Reference Landauer and Dumais1997), in which meaning is represented by arbitrary symbols (e.g., binary digits) that do not retain any of the properties of the sensorimotor states that gave rise to them (see review in Markman & Dietrich Reference Markman and Dietrich2000). Similar to “analogical” and “modal” accounts, embodied models propose that meaning is constituted by (or “grounded in”) sensorimotor information, such as that furnished by facial expressions. Since their advent, embodied models have been challenged on empirical grounds. Regarding emotion, for example, an extensive study revealed that people with bilateral facial paralysis stemming from Moebius syndrome can categorize the facial expressions of others without difficulty (Bogart & Matsumoto Reference Bogart and Matsumoto2010). Moreover, the ability of subjects to express facial expressions does not influence accuracy. Similarly, in semantics, it remains unclear whether lesions of action-related regions of the brain lead to impairments regarding the semantic-conceptual knowledge of action: Patients suffering from deficits due to brain lesions can show impairments in the way they use certain objects but, revealing the abilities of a “disembodied” conceptual system, can nonetheless name those objects, or even recognize the pantomimes associated with them (Mahon & Caramazza Reference Mahon and Caramazza2008).
In response to such challenges, embodied approaches now propose that sensorimotor information plays a more humble role: In the target article, Niedenthal et al. portray sensorimotor information not as that which constitutes meaning, but merely as auxiliary information that is “referred to” when semantic processing is challenged, as when categorizations are difficult to make. In these accounts, sensorimotor information is demoted to a kind of supplementary information that, just like any kind of information, can at times facilitate or interfere with semantic processing. When the strategy of embodied simulation fails, it is the conceptual system that arrives at the scene to solve the categorization, much like a deus ex machina.
The question now is whether the conceptual system is always necessary for categorization. From one standpoint, the answer is yes: Until a token facial expression is typed as a member of a given category (an inherently conceptual process), the organism cannot know (consciously or unconsciously) the meaning of the observed expression. If the conceptual system must always be at the scene, then what is its nature and how does it solve the symbol grounding problem? Embodied approaches were expected to answer this by demonstrating, for example, that disrupting the sensorimotor states in which a conceptual process is grounded will not just modulate the process, but also eradicate it. It seems that, apart from modulations that could be construed as priming or its opposite (i.e., interference), there is no evidence of conceptual processing requiring sensorimotor states. Apart from the aforementioned neurological evidence, this is evident with respect to abstract concepts (e.g., that of if and justice) and everyday phenomena such as imageless thought (e.g., a state of uncertainty or readiness; cf. Hilgard Reference Hilgard1987). Regarding imageless thought, Woodworth (Reference Woodworth1915) concluded, “I know of no reason in neurology or psychology for supposing that the elements of conscious content are contributed solely by the sensory receiving centers” (p. 137).
A way forward? Identifying the limitations of “disembodied” processing
Perhaps a critical next step in theory-building for embodied approaches is to identify the unique contribution(s) of embodied simulation. What is it that a disembodied conceptual system is incapable of doing with respect to semantic processing or the categorization of smiles? Emotion research has broached this topic by revealing some limits of language, a form of propositional processing. Evidence suggests that linguistic processing does not lead to the same kinds of emotional effects and forms of affective learning as modal stimuli (LeDoux Reference LeDoux1996). Olsson and Phelps (Reference Olsson and Phelps2004) demonstrated that, through vicarious classical conditioning, subjects can acquire a learned fear response toward a subliminal stimulus by perceiving someone else receive a shock after being presented with that stimulus. However, simply telling subjects about the contingency between stimuli did not lead to this form of vicarious conditioning to subliminal stimuli. This suggests that the phylogenetically old systems mediating fear conditioning may not understand language, though they can process the meaning of basic perceptual events: “classical conditioning and observational learning … might be supported by an evolutionarily old system that predates the emergence of language” (Olsson & Phelps Reference Olsson and Phelps2007, p. 1099).
That language fails to yield the vicarious conditioning afforded by observational learning may be because language comprehension yields “faint” activations of analogical, sensorimotor-like symbols (Boroditsky & Ramscar Reference Boroditsky and Ramscar2002; Zwaan Reference Zwaan, DeVega, Glenberg and Graesser2008) or, at odds with embodied accounts, because linguistic-semantic knowledge is simply distinct from sensorimotor knowledge. Regarding the pioneering investigation by Niedenthal et al., the next advancement may be to isolate that which only embodied simulations can achieve. It may be that aspects of social mimicry and empathy require embodied simulation, or that the three smile-categorization mechanisms operate differently with respect to subliminal or supraliminal stimuli, with the conceptual deus ex machina being capable of arriving at the scene only when the theater lights of consciousness are on.