?Smart? mechanisms emerging from cooperation and competition between modules

Abstract

Three sets of experiments are described. The first set concerns the detection of the 3-D possibility of shaded 2-D block patterns. The data indicate that the human perceptual system is able to do this above chance level, but in a specific and restricted way, which suggests the possibility of a module devoted to it. The second set concerns the determination of the 3-D orientation of coloured 2-D patterns. The data show that the human perceptual system uses physical constraints of colour mixing in doing this. Again, the mechanism seems to have characteristics suggesting modularity. The third set is about the detection of skewed symmetry in dot patterns. Skewed symmetry can be used as a source of information about the slant and tilt of a surface with bilateral symmetry present on it. Therefore, a module for recovering this information might exist. All these empirical data show that Fodor's criteria of modularity can be used to test hypotheses about visual mechanisms recovering 3-D information from 2-D inputs. But this does not imply that the results prove modularity and cannot be interpreted otherwise. As an exercise the same story about these experiments is told with the use of jargon from ecological realism and connectionist approaches. It is concluded that some of the approaches to some perceptual mechanisms are not as divergent as they might seem. The main theme underlying modular, connectionist, and ecological approaches is the avoidance of central intelligence agencies by the incorporation of physical constraints. Within this broad framework different questions can be asked and answers attempted that may depend on personal taste.

Skewed symmetry: A nonaccidental property used to perceive visual forms.

“…In Marr's (1982) view, these and other aspects of depth (e.g., distance to the viewer) could then be represented in a metric surface representation, called a 2 1/2-dimensional sketch. Doubts about the possibility and usefulness of such an accurate depth map and the so-called modules to compute it have been expressed before, both in computer vision (e.g., Barnard, 1985; Lowe, 1987) and in perceptual psychology (e.g., Todd & Akerstrom, 1987; Wagemans, 1988, 1990). The data gathered here suggest that similar doubts are warranted for this specific case of skewed symmetry as well.…”

Perceptual use of nonaccidental properties.

“…In the second set of four experiments, which are described here for the first time, an attempt is made to test whether subjects can infer the 3-D orientation from SS. In a sense, the latter experimental task can be regarded as a shape-from-skewedsymmetry problem, analogous to other shape-from methods such as stereo, motion, shading, and texture (for a review of the computational and psychological research of these so-called "modules", see Wagemans, 1988Wagemans, , 1990. The experiments about "ss as NAP" are summarized in Part II, whereas those focused on "Shape from SS" are described in Part III.…”