Authors:

Duncan, J.

Reference:

Year of publication:

1989

CBU number:

2332

Abstract:

This commentary concerns the binding problem in parallel, distributed perceptual models: Given simultaneous activation of several distributed object representations, what means is used to show which parts belong together? As an example, what means is used to bind together the colour, shape etc. of individual objects in a visual scene? A switch to serial processing across locations is the common solution, but this is questioned here on several grounds. First, search data show there can be parallel processing of conjunctions of visual attributes. Second, common location may not be the only tag binding together an object's different attributes; other grouping cues also influence the perception of feature conjunctions. A possible direction is suggested for parallel solutions, based on the fact that visual neurons rarely behave as pure feature detectors. Instead activation indicating the presence of any particular feature also carries an indication of several other features paired with it - including but not restricted to location.