Pylyshyn makes a convincing case that early visual processing is cognitively impenetrable, and although I question the utility of binary oppositions such as penetrable/impenetrable, for the most part I am in agreement. The author does not provide explicit designations or denotations for the terms penetrable and impenetrable, which appear quite arbitrary. Furthermore, the use of focal attention smacks of an homunculus, and the account appears to slip too easily between the perceptual, the cognitive, and the neurophysiological.

Signal detection theory (SDT) is best known as a method for separating sensitivity from bias. If sensitivity reflects early sensory processing and bias later cognition, then SDT can be use to study penetrability by asking whether cognitive manipulations affect sensitivity. This assumption is too simple, but SDT can nonetheless be helpful in developing specific methods of how sensory and cognitive information combine. Two such approaches are described.

Discussions of the relationship between perception and cognition often proceed without a definition of these terms. The sensory-modality specific nature of low-level perceptual processes provides a means of distinguishing them from cognitive processes. A more explicit definition of terms provides insight into the nature of the evidence that can resolve questions about the relationship between perception and cognition.

Pylyshyn argues that early vision is cognitively impenetrable, and therefore – contrary to knowledge-based theories of perception – that perception is noncontinuous with cognition. Those processes that are included in “early vision,” however, represent at best only one component of perception, and it is important that it is not the component with which most knowledge-based theories are concerned. Pylyshyn's analysis should be taken as a possible source of refinement of knowledge-based theories of perception, rather than as a condemnation of them.

Pylyshyn's model of visual perception leads to problems in understanding the nature of perceptual experience. The cause of the problems is an underlying lack of clarity about the relation between the operation of the subpersonal vision module and visual perception at the level of the subject or person.

Basic processes of perception should be cognitively impenetrable so that they are not prey to momentary changes of belief. That said, how does low level vision interact with knowledge to allow recognition? Much more needs to be known about the products of low level vision than that they represent the geometric layout of the world.

Among Pylyshyn's most important questions is determining the boundaries of early vision. A simple stimulus illustrates that, in addition to the dominant percept, most observers can perceive alternative interpretations of 3-D surface layout only after provided with suggestions. These observations may indicate that cognitive influences reach the stages of visual processing where 3-D surface configurations are resolved.

Although some types of cognition may not affect early vision, there is ample evidence that other types of cognition do. Evidence indicating that early vision is penetrable by direct manipulation of viewers' perceptual intentions and by knowledge of the structure of familiar objects is reviewed, and related to both the Pylyshyn target article and Fodor (1983).

How can the impenetrability hypothesis be empirically tested? We comment on the role of signal detection measures, suggesting that context effects on discriminations for which post-perceptual cues are irrelevant, or on neural activity associated with early vision, would challenge impenetrability. We also note the great computational power of the proposed pre-perceptual attention processes and consider the implications for testability of the theory.

Early vision provides general information about the environment that can be used for motor control or navigation and more specialized information that can be used for object recognition. The general information is likely to be insensitive to cognitive factors, but this may not be entirely true for the information used in model-based recognition.

The commentator agrees with Pylyshyn's most general claims but sees problems with the more specific proposals about where the boundary between early vision and later processing might lie. The boundary cuts across current models of identification. Limitations in current research on scenic context effects preclude firm conclusions. High-level vision will benefit more from integrative work than from premature analysis.

Pylyshyn's concentration on form perception to demonstrate that early vision is cognitively impenetrable neglects that color perception is also part of early vision. Thus, the finding of Duncker (1939), Bruner et al. (1951), and Delk and Fillenbaum (1965) that the expected color of objects affects how they are perceived challenges Pylyshyn's thesis.

Pylyshyn acknowledges that cognition intervenes in determining the nature of perception when attention is allocated to locations or properties prior to the operation of early vision. I present evidence that scale perception (one function of early vision) is cognitively penetrable and argue that Pylyshyn's criterion covers not a few, but many situations of recognition. Cognitive penetrability could be their modus operandi.

Pylyshyn's target article argues that perception is not inferential, but this is true only under a narrow construal of inference. A more general construal is possible, and has been used to provide formal theories of many visual capacities. This approach also makes clear that the evolution of natural constraints need not converge to the “veridical” state of the world.

Pylyshyn claims that if a system is cognitively penetrable, its function depends in a semantically coherent way to the organism's goals and beliefs. He rejects evidence of attentional modulation observed in neurons within the visual system, claiming that any modulation seen is not logically related to goals and behavior. I present some of this evidence and claim that it is connected in exactly the way Pylyshyn requires and thus it refutes his main thesis.

Expert perceivers may learn more than just where to apply visual processing, or which part of the output from the visual system to attend to. Their early visual system may be modified, as a result of their specific needs, through a process of early visual learning. We argue that this is, in effect, a form of long-term, indirect cognitive penetration of early vision.

Pylyshyn circumvents an even more fundamental question: Are the mechanisms of visual perception accessible to the theoretician? Neurophysiology, computer modeling, and psychophysics, as well as his definitions of visual phenomena suggest that he has asked an unanswerable question.

It is argued that the alleged cases of cognitive penetration of visual modules actually arise from the integration of information among different modules. This would reflect a general computational strategy according to which constraints to a particular module would be provided by information coming from different modules. Examples are provided from the integration of stereopsis and occlusion and from computation of motion direction.

Pylyshyn's effort in establishing the cognitive impenetrability of early vision is welcome. However, his view about the role of attention in early vision seems to be oversimplified. The allocation of focal attention manifests its effect among multiple stages in the early vision system, it is not just confined to the input and the output levels.

We offer an ecological (Gibsonian) alternative to cognitive (im)penetrability. Whereas Pylyshyn explains cognitive (im)penetrability by focusing solely on computations carried out by the nervous system, according to the ecological approach the perceiver as a knowing agent influences the entire animal-environmental system: in the determination of what constitutes the environment (affordances), what constitutes information, what information is detected and, thus, what is perceived.