Spatial mechanisms in the recovery of visual feature conjunctions

Abstract

Recent neurophysiological and behavioural evidence suggests that in both early visual processing and memory many of the component features of an object are independently represented in the visual system. The unity of visual experience appears to be dependent upon the recovery of feature conjunctions. The mechanism(s) responsible for re-establishing appropriate feature combinations has yet to be fully explicated but it appears possible that the common spatial location of features may provide a basis for a solution to the 'feature binding problem'. A number of writers have postulated that spatial localization provides a mechanism for recombining features by offering a means of cross-referencing independent spatially tagged feature maps (Isenberg, Nissen & Marchak, 1990; Johnston & Pashler, 1990; Nissen, 1985; Van der Heijden, 1993). The present study investigated localization dependence in inter-feature tasks using a partial report paradigm in which subjects were presented with multi-clement displays and post-cued to report one or more target features of the cued item. The first two experiments replicated Nissen (1985, Experiment 2). Following Nissen's analysis the data generally reproduced her findings of stochastic independence in location-cued shape and colour report contrasting with localization dependence in colour-cued shape and location report.
However, when more sensitive analytical techniques were applied several anomalous results emerged which suggested that the Nissen paradigm may be incapable of producing unambiguous findings. Subsequent experiments used alternative experimental procedures which avoided the ambiguities inherent in the Nissen paradigm and provided a more direct investigation of the localization dependence hypothesis. Experiments 3 - 6 provided converging evidence that in inter-feature tasks access to the target feature is spatially mediated.
Experiments 7 and 8 examined the generality of the spatial feature binding mechanism and obtained evidence that its operation is not confined to the recovery of shape-colour conjunctions and extends to a variety of feature combinations. Experiments 9 and 10 examined the representation and retention of multifeatured items over longer retention periods. The
results of Experiment 9 suggested that while spatial location continues to provide a feature binding code in STM, in the absence of a spatially distributed display, feature conjunctions may be temporally coded in memory. Experiment 10 revealed a pronounced spatial adjacency effect in STM. Further analysis indicated that both a spatial feature binding mechanism and a somewhat less efficient verbal strategy operate in parallel to retain feature conjunctions in shortterm memory (STM). It was concluded that the findings of the present study generally provide convergent evidence that while verbal and temporal codes may support the retention of feature conjunctions, spatial localization appears to provide the principal feature binding mechanism. Implications of the present findings for current theories of feature integration were discussed and a theoretical model of the short term retention of multidimensional stimuli was developed in which both the organization of material in visual STM and the maintenance of feature conjunctions is specified.