homeresearch › phd

Basic Features in Vision

An analysis of constraints on perceptually relevant sensory properties

[basic features picture]

PhD in Cognitive Science.
Awarded by the Ecole des Hautes Etudes en Sciences Sociales (Paris), 6 October 2005.
Mention: très honorable avec félicitations du jury.

Roberto Casati (CNRS)

Jérôme Dokic (EHESS)
Pierre Jacob (CNRS)
Pascal Mamassian (CNRS)
Zenon W. Pylyshyn (Rutgers University)

Hosting lab
Institut Jean Nicod (UMR 8129)

This project was supported by a 3-year individual fellowship (allocation de recherche) from the French Ministry of Education (2001-2004) and by a 1-year grant from the European Network of Excellence ENACTIVE (2005) (IST-2002-002114).


This work addresses a theoretically contentious issue in perceptual science, namely the characterization of "basic visual features" (i.e., properties of the visual stimulation that are assumed to be the elementary constituents of the functional input of visual perception). The goal of this work is twofold.
On the one hand, I articulate the traditional constraints on feature basicness that rely on merely internalist criteria. "Primitiveness constraints" - just to mention the most important example - represent a paradigmatic case of basicness criteria that rely on assumptions traditionally drawn from a merely internalist account of the structure and functioning of perceptual systems.
On the other hand, I contrast this class of constraints with constraints based on adaptive criteria, i.e. criteria that define what a basic feature is on the basis of a match between an organism's adaptive needs and the structure of its sensory environment. I review, in particular, some case studies of perceptual abilities in which regularities in the sensory environment seem to play a major role in the definition of feature basicness, like perceptual shunt mechanisms.

Perceptual shunt mechanisms (Keil, 2000; Keil et al., 2002) can be described as perceptual strategies allowing the solution of relatively complex perceptual problems by relying on low-level patterns that are robustly correlated with high-level properties within specific environmental contexts. These correlations between low-level patterns and high-level properties (or corresponding natural constraints embodied in perceptual systems) can account for several cases of apparent smart behavior in organisms with limited perceptual and cognitive resources. It is in virtue of these contingent (although statistically robust) correlations that seemingly complex categorization tasks or high-level visual routines can be efficiently accomplished by relatively simple organisms. For example, a bee's ability to solve complex 3D topological tasks (Chen et al., 2003; Pomerantz, 2003) can be accounted for in terms of a sensitivity to relatively simple sensory configurations that bear a statistically reliable correlation with the considered topological properties within their environmental niche.

The interest in the study of cases of perceptual abilities strongly dependent on the contingent structure of the perceptual environment is motivated by two main reasons.
On the one hand, such cases allow to revise the traditional criteria for feature basicness, by drawing the attention of researchers on properties and stimulus patterns that have been often disregarded as possible candidates for basic perceptual features.
On the other hand, basic features defined on the basis of adaptive constraints can help dissolve a number of ill-posed problems in perceptual science, which assume that perceptual competence has to be explained without taking into account regularities of the environment in which an organism is embedded.



Valid XHTML 1.0 Transitional :: Valid CSS :: Powered by WikkaWiki