The motor system might well use deictic strategies when subjects learn a new task. However, its not clear that Ballard et al. show this. Multiple eye-fixations may have little to do with deixis and more to do with the unfamiliarity of the task. In any case, deixis does not entail embodiment, since a disembodied Cartesian brain could use deictic strategies.

Ballard et al. show how control structures using minimal state can be made flexible enough for complex cognitive tasks by using deictic pointers, but they do so within a specific computational framework. We discuss broader implications in cognition and memory and provide biological evidence for their theory. We also suggest an alternative account of pointer binding, which may better explain their limited number.

Deictic pointers allow the nervous system to exploit information in a frame that is centered on the object of interest. This processing may take place in visual or haptic space, but the information processing advantages of deictic pointing can also be applied in abstract spaces, providing the basis for analogical transfer. Simulation and behavioral results illustrating this progression from embodiment to abstraction are discussed.

Following the “modularity” orthodoxy of some years ago, it has traditionally been assumed that there is a clear and obvious separation between perception and cognition. Close examination of this concept, however, fails to reveal the join. Ballard et al.'s contention that the two “cannot be easily separated” is consistent with nonmodular views of the way that symbol grounding might be achieved in situated systems. Indeed, the traditional separation is viewed as unhelpful.

It is unlikely that Ballard et al.'s embodiment theory has general applicability to cognition because it is based on experiments that neglect the importance of meaning, purpose, and learning in cognitive tasks. Limitations of the theory are illustrated with examples from everyday life and the results of recent experiments using cognitive and visuomotor tasks.

Embodiment, the explicit dependence of cognition on the properties of the human body, is the foundation of contemporary cognitive science. Ballard et al.'s target article makes an important contribution to the embodiment story by suggesting how limitations on neural binding ability lead to deictic strategies for many tasks. It also exploits the powerful experimental method of instrumented virtual reality. This commentary suggests some ways in which the target article might be misinterpreted and offers other cautions.

This commentary centres around the system of human visual attention. Although generally supportive of the position advocated in the target article, we suggest that the detailed account overestimates the capacities of active human vision. Limitations of peripheral search and saccadic accuracy are discussed in relation to the division of labour between covert and overt attentional processes.

Questions about the validity of eye fixations in the blocks task as a memory indicator are discussed. Examples from reading research illustrate the influence of extraneous factors.

Ballard et al.'s is a plausible and useful model. Leaving aside some unnecessary constraints, the model would probably be valid through a wider gamut of interactions between the self and the environment than the authors envision.

Ballard et al. claim that fixations bind variables to cognitive pointers. I comment on three aspects of this claim: (1) its contribution to the interpretation of indexical language; (2) empirical support for the use of very few deictic pointers; (3) nonetheless, abstract pointers cannot be taken as prototypical cognitive representations.

Deictic coding offers a useful model for understanding the interactions between the dorsal and ventral streams of visual processing in the cerebral cortex. By extending Ballard et al.'s ideas on teleassistance, I show how dedicated low-level visuomotor processes in the dorsal stream might be engaged for the services of high-level cognitive operations in the ventral stream.

Ballard et al. model eye position as a deictic pointer for spatial perception. Evidence from research on gaze control indicates, however, that shifts in actual eye position are neither necessary nor sufficient to produce shifts in spatial perception. Deictic context is instead provided by the interaction between two deictic pointers; one representing actual eye position, and the other, intended eye position.

Ballard et al.'s principle of deictic coding as exemplified in the analysis of fixation patterns relies on a functional dichotomy between foveal and extrafoveal vision based on the well-known dependency of spatial resolution on eccentricity. Experimental evidence suggests that for processes of pattern learning and recognition such a dichotomy may be less warranted because its manifestation depends on the learning state of the observer. This finding calls for an explicit consideration of learning mechanisms within deictic coding schemes.

The embodied AI paradigm represents a distinct shift toward an ecological perspective on intelligent behavior. I outline how Ballard et al. have made a promising advance in expanding the seat of intelligence to include sensory and motor systems, but they have not gone far enough. Sustained growth toward truly generalizable accounts of intelligent systems will also require expanding the locus of intelligence to include the environmental structure participating in intelligent behavior.

Ballard et al. neatly demonstrate the tradeoff between memory and externalised serial deictic functions which help the agent achieve economy. However, the target article represents an implementation which does not seem to reveal the hidden level of cognitive control enabling tasks, contexts, and the agent's own needs to specify and recruit information from the visual layout.

The target article makes a compelling case for the idea that agents rely on the world for external memory in fast-paced perceptual tasks. As I argue, however, agents also rely on the external environment for computational hardware that helps to keep cognitive computations tractable. Hence the external world provides not only memory for computations involving perceptual system actions, but it provides domain-level actions that figure in cognitive computations as well.

Traditional cognitivism treats a situated agent's point of view in terms of deficit egocentrism. Can Ballard et al.'s framework remedy this characterization? And will its fusion of computational and enactivist explanations change assumptions about what cognition is? “Yes” is suggested by considering human infants' developing spatial knowledge, but further questions are raised by analysis of their robot counterparts.

Evolutionary pressure selects for the most efficient way of information processing by the brain. This is achieved by focussing neuronal processing onto essential environmental objects, by using focussing devices as pointers to different objects rather than reestablishing new representations, and by using external storage bound to internal representations by pointers. Would external storage increase the capacity of cognitive processing?

Eye movements play a variety of roles in perception, cognition, and communication. The roles are revealed by the duration of fixations reflecting the quality of processing in the first line. We describe possible roles of eye fixations in different temporal diapasons. These forms of processing may be specific to sensorimotor coordinations. Any generalization to other domains should be cautious.

What is encoded in working memory may be a content-addressable pointer, but a critical portion of the information that is addressed includes the motor information to achieve deictic reference in the environment. Additionally, the same strategy that is used to access environment information just in time for its use may also be used to access long-term memory via the pre-frontal cortex.