What Do We Learn From Binding Features? Evidence for Multilevel Feature Integration Lorenza S. Colzato Leiden University Antonino Raffone University of Sunderland and RIKEN Brain Science Institute Bernhard Hommel Leiden University Four experiments were conducted to investigate the relationship between the binding of visual features (as measured by their aftereffects on subsequent binding) and the learning of feature– conjunction probabilities. Both binding and learning effects were obtained, but they did not interact. Interestingly, (shape– color) binding effects disappeared with increasing practice, presumably be- cause of the fact that only 1 of the features involved was relevant to the task. However, this instability was only observed for arbitrary, not highly overlearned combinations of simple geometric features and not for real objects (colored pictures of a banana and strawberry), where binding effects were strong and resistant to practice. These findings suggest that learning has no direct impact on the strength or resistance of bindings or on speed with which features are bound; however, learning does affect the amount of attention particular feature dimensions attract, which again can influence which features are considered in binding. Keywords: feature integration, learning, binding problem, attentional set, event file Considerable evidence suggests that cortical networks encode the external environment in a distributed fashion. A striking example of spatially distributed coding in cortical information processing is given by the primate visual cortex, processing visual event features in parallel in numerous cortical maps (Cowey, 1985; Felleman & van Essen, 1991). This coding scheme also applies to events in the auditory and other sensory modalities and to multimodal event processing. As handy as distributed coding may be, it creates numerous so-called “bind- ing problems,” that is, difficulties in relating the codes of a given entity or processing unit (e.g., visual object) to each other. To resolve these problems, the brain needs some sort of integration mechanism that binds together the distributed codes belonging to the same event, while keeping these codes sepa- rated from codes for other events (Treisman, 1996). Mechanisms of Feature Integration At a neural level, a theoretical solution to the binding problem may be given by high-order cardinal cells (Barlow, 1972), onto which signals from neurons coding for the to-be-bound features converge. However, given the high variability of objects belonging to a given category in terms of their instances and retinal projec- tions, as well as the numerous ways in which discrete features can be potentially combined, the exclusive reliance on convergent mechanisms would ultimately lead to a combinatorial explosion and is therefore not plausible. Another potential solution to the binding problem is given by cell (neural) assemblies or sets of tightly connected neurons, the identity of the assembly being defined in terms of higher firing rates or coactivation of the participating neurons (Amit, 1995; Braitenberg & Schu ¨z, 1991; Hebb, 1949). In this representational scheme, individual neurons encode for simple features, and the associative connections between these neurons enable pattern encoding and completion within the assembly. This solution avoids the combinatorial explosion problem implied by cardinal cells and, thus, seems to be well suited for arbitrary, frequently changing feature combinations. At a behavioral level, one way to study feature binding mech- anisms is to put processing systems under conditions that render proper integration difficult or impossible and then to look for the creation of incorrect bindings or “illusory conjunctions” (Treisman & Schmidt, 1982). Another way is to search for aftereffects of feature integration, that is, for side effects of created feature bindings on later performance. In a seminal study along these lines, Kahneman, Treisman, and Gibbs (1992) presented partici- pants with two displays in a sequence, a brief multiletter prime Lorenza S. Colzato and Bernhard Hommel, Department of Psychology, Cognitive Psychology Unit, Leiden University, Leiden, the Netherlands; Antonino Raffone, Theoretical and Applied Simulation Laboratory, Uni- versity of Sunderland, Sunderland, United Kingdom, and Laboratory of Perceptual Dynamics, RIKEN Brain Science Institute, Wako, Saitama, Japan. Lorenza S. Colzato and Bernhard Hommel are members of the Exper- imental Psychology Research School. We thank Raymond Klein, Bruce Milliken, and Derrick Watson for comments on a draft of the manuscript. Correspondence concerning this article should be addressed to Bernhard Hommel, Department of Psychology, Cognitive Psychology Unit, Leiden University, Postbus 9555, 2300 RB Leiden, the Netherlands. E-mail: hommel@fsw.leidenuniv.nl Journal of Experimental Psychology: Copyright 2006 by the American Psychological Association Human Perception and Performance 2006, Vol. 32, No. 3, 705–716 0096-1523/06/$12.00 DOI: 10.1037/0096-1523.32.3.705 705