The special issue of Psychonomic Bulletin and Review, in conjunction with the digital #symbodiment event, represents an effort to take stock of the “embodiment vs. symbols” debate that has garnered an increasing amount of attention in the field. In this commentary, I present a few thoughts about the successes and failures of the embodied research program, and offer some thoughts on the road forward.
Although embodied and symbolic approaches are often portrayed as being in strong opposition (see the exchange between Mahon and Glenberg in 2015), the perspectives on display in this special issue paint a different picture. For better or worse, there are many respects in which the embodied and symbolic perspectives are quite similar (see Barsalou’s article in the #symbodiment special issue).
Consider the case of language comprehension, a research area in which the embodiment vs. symbols debate has been particularly active over the past 15 years. The early embodied work in this area was aimed at building the case that the representational format of existing (symbolic) theories of comprehension could not explain the perceptual/motor compatibility effects that were observed. The point of distinction between the embodied and symbolic theories was representational format; outside of this, the embodied approaches had much in common with the extant symbolic models – proponents of both approaches would agree that syntactic processing, background knowledge, and so on, were key elements of the comprehension process.
I think it is reasonable to say that much of the debate between embodied and symbolic accounts in the ensuing years has been also been about representational format. Martin argues in the #symbodiment special issue that debates about representational format are unlikely to lead to progress in the field, since we have little understanding of both the representational format of the brain, and the methods that would be required to determine what that format actually is. I agree with Martin’s assessment. If the debate between embodied and symbolic theories is a debate about representational format, it is unlikely to be effective in pushing the field forward.
Evaluating the embodiment research program on its own terms, I would say that it has been partially successful. On the positive side, research inspired by embodiment has shown that systems of perception, action planning, and emotion are more involved in many types of cognitive processing than had previously been thought (as the literature reviews provided in the #symbodiment special issue amply show). Additionally, like the visual world paradigm before it, the study of embodiment broadened the methodological palate of cognitive scientists, and served to highlight the extent to which the dynamics of different task responses could be exploited to study cognitive processes. Finally, and importantly, the theorizing surrounding the embodied position served to highlight the weakness of the case for the standard symbolic model of cognition. As Barsalou notes in the special issue, the fact that so many researchers take the symbolic model as given obscures both the fact that the evidence for symbolic representations and symbol system-type processing is weak, and the fact that such theories have serious shortcomings as accounts of cognition. The empirical case for symbolic approaches to cognition is no better than – and, all things considered, may well be worse than – the case for embodiment. To echo Barsalou, it is amazing that the symbolic approach should be seen as the default view of the field under such circumstances. These contributions point to the success of the embodiment program.
The major negative to the embodied program is that it has largely failed to move much further than to demonstrate that perceptual, motor, or emotional factors contribute to cognitive processing. For example, as noted above, embodied accounts of language comprehension could be interpreted as modified versions of extant symbolic theories – the framework of processing remained the same, but the representational content was tweaked to be sensorimotor rather than symbolic or propositional (Chemero made a similar argument).
That embodied theories should lean on symbolic theories for their structure is understandable. At the time that the embodied perspectives were developing, there were many fleshed-out symbolic models of comprehension (e.g., Kintsch’s Construction-Integration model), and standard symbolic cognitive science had developed a tool kit of constructs and devices that could be invoked to play an explanatory role in a broad range of theories – priming, working memory, attention, and the like. The problem here is not that embodied approaches to language (or, other phenomena) borrowed basic assumptions from the extant symbolic theories. Many of the important ideas from models such as Kintsch’s (e.g., knowledge is important to the comprehension process) are intuitive and likely to be correct in some sense, regardless of whether cognition turns out to be “symbolic” or “embodied” (or, none of the above). A wholesale rejection of these ideas seems to be unwarranted. Rather, the problem is that by implicitly or explicitly taking on the tool kit of constructs from the symbolic approach, the embodied approach also took on the problems inherent in these approaches to understanding cognitive processes (van Orden and colleagues discuss the dangers of this “took kit” approach).
It is useful to clarify that my comments about the successes and failures of the embodied research program are aimed at what might be called “standard” embodiment: studies of the role of perception and action in language processing, memory retrieval, and other standbys of the cognitive literature, with adult research participants. I have done so because this is the literature that tends to feature prominently in the embodiment/symbolic debates (such as the exchange between Mahon and Glenberg and the reviews provided in this special issue), and is probably the most widely known work that falls under the embodiment umbrella.
In the remainder of this commentary, I consider the future of embodiment, and the role that embodied theorizing might play in developing our understanding of cognition. In doing so, I discuss two elements of the broader embodied research program that are conspicuous in their absence from the special issue: radical embodied cognition (REC) as pursued by Chemero, Clark, and Wilson & Golonka, and development, as pursued for example by Smith.
Radical embodied cognition differs from both “standard” embodiment and symbolic accounts in that it pursues a non-representational approach to cognition (see Chemero and Wilson & Golonka for discussion; but see Clark for the suggestion that REC need not be non-representational). The distinction between REC and representational approaches can be illustrated by examining different explanations for how outfielders catch fly balls. As discussed by Wilson and Golonka, standard representational accounts of this behavior would suggest that the outfielder uses internal representations (such as internal physics) to calculate the trajectory of the ball and move to the appropriate location. REC accounts suggest that outfielders use the dynamic, moment-by-moment coupling between their movement and the resulting changes in their optic array to converge on the location of the ball.
Radical embodied cognition, and the Gibsonian perspective from which it springs, is a controversial position, and it is beyond the scope of this commentary to discuss or resolve the relevant controversies. Nonetheless, I think REC can play an important role as the field continues to think about concepts, meaning, and embodiment. REC involves a commitment to the idea that behavior needs to be understood in context. Paraphrasing Wilson and Golonka, in each context, there is both a task at hand and information present in the environment (as well as in the history of the organism in those kinds of environments). Our goal is then to understand how the information is used to resolve the task within that context.
As discussed by Yee and Thompson-Schill in the #symbodiment special issue, it is becoming increasingly clear that conceptual knowledge (and performance on conceptual tasks) is highly context-dependent. Van Orden and colleagues argued that the standard symbolic approach to cognition is ill-suited to capture this sort of context-dependency, as it is forced to develop an ever-expanding set of constructs to explain the context-dependent changes in behavior. REC and related dynamic systems approaches, with their orientation toward the dynamic coupling of the organism and current environment, suggest a more unified approach to problems of context dependency.
Development is an important, and perhaps indispensable, component of the REC approach. In cases where REC would seem to be challenged to explain particularly complex adult behavior (e.g., language use), a developmental perspective may provide a more profitable path forward. For instance, it may be more straightforward to develop an REC account of children’s developing language (or conceptual) abilities and then extend this forward to adulthood, than to start by attempting to model adult language (or concept-related) behavior. The classic work by Thelen and colleagues on the A-not-B error in young children provides a sterling example of how an REC approach can be applied to developmental phenomena. Thelen and colleagues demonstrated that the A-not-B task, once taken to reflect the development of conceptual knowledge (in this case, object permanence), can be understood in terms of the information presented to the child, the child’s history of reaching in the task, and the characteristics of the reaching task itself. Smith and Jones argued in 1993 that the variability and flexibility that children show in their development of vocabulary and conceptual knowledge points to the need to move away from the standard symbolic approaches to cognition and toward a more contextual, dynamic approach. Doing so will not necessarily be easy; as van Orden and colleagues note, taking a dynamic perspective seriously will likely require us to develop a new methodological toolbox.
It is my sense that the contextual variability documented by Yee and Thompson-Schill in this issue points to one of the most fundamental issues facing cognitive science – how are we to understand the system that gives rise to our flexible, adaptable, intelligent behavior? As discussed by van Orden and colleagues and by Smith and Jones, and others, the standard symbolic approach to cognition (and, by extension, the embodied approaches that lean heavily on the symbolic approach) is ill equipped to explain the contextual nature of behavior, and therefore ill equipped to address this issue. Radical embodied cognition provides a promising perspective from which to tackle problems of contextual, flexible behavior. There is clearly much work to do to develop the REC approach, and whether (and how) REC can scale up to handle complex behavior such as language remains an open question. Nonetheless, I believe that the pursuit of this direction will have put us on a fruitful path.