Focused attention in the perception and retrieval of multidimensional stimuli. Treisman, A. Perception & Psychophysics, 22(1):1-11, 1977.
Focused attention in the perception and retrieval of multidimensional stimuli [pdf]Paper  abstract   bibtex   
This paper reports some further experiments on successive matching of multidimensional stimuli in which the correct conjunctions of features must be specified; it also modifies and extends the model proposed earlier by Treisman, Sykes, and Gelade (1977). The results obtained in the previous experiment were replicated despite a change from fixed to varied targets, and from spatial to temporal separation of the targets, thus extending their generality. The modified conjunction-matching model proposes that the subject progressively narrows his focus of atten-tion, first to deal with a single display stimulus at a time and finally, if necessary, to a single target. The model was tested quantitatively by simulating the separate stages hypothesized to underlie conjunction-matching in some simpler, single-attribute-matching conditions, and using the differences between latencies in these conditions to predict differences between latencies in the appropriate conjunction-matching conditions. The results were consistent with the general hypothesis that one important role of focused attention and serial processing is to integrate separable attributes or features into the correct conjunctions, which correspond to the objects actually presented or stored. Most of the objects, people, and scenes we perceive produce complex, multidimensional, changing patterns of stimulation on the retina. At least some of their attributes appear to be registered by in-dependent neural channels, specializing in different aspects, such as orientation, color, spatial frequency, brightness. Analysis of complex stimuli into more elementary features may be logically necessary in order to allow generalization across instances of the many categories we form and, at the same time, dis-crimination between them. But this immediately raises the question of how the component properties are resynthesized into the correct compounds, so that we correctly see the shirt as blue and the trousers as gray, for example, rather than the reverse. Treisman, Sykes, and Gelade (1977) offered some preliminary suggestions and findings relating to this problem, in the context of visual search and visual matching. This paper reports some further investiga-tions of the matching task, which extend the general-ity of the findings; it then attempts to develop and test the model we proposed earlier. There were two main areas of interest in the previous paper: (1) The first was the hypothesis that separable dimensions or features [i.e., those that are processed by independent analyzers (Garner, 1974)] are perceptually integrated into the correct com-pounds by means of focal attention and serial processing of information from different spatial locations. Features which are registered in the same location and temporal interval, i.e., within the same central focus of attention, can then be coded as belonging to the same object. This implies that parallel processing and divided attention to multiple inputs should be possible only when either of two conditions is met: (a) the inputs can be distinguished by differences in a single feature, or (b) they vary along different dimensions, so that erroneous inter-changes of features leading to illusory conjunctions cannot occur. The first experiment we reported tested the prediction that serial processing would be re-quired in a visual search task if the target differed from non targets only in its particular combination of features, while if it could be distinguished by a single feature processing would be parallel. The results were consistent with this hypothesis. (2) The second problem we began to explore was how conjunctions, once perceived, are held in work-ing memory and matched in a recognition task. We noted three possible methods. The first depends on a unitization process (cf. Laberge, 1973), in which higher level units are formed to encode directly the formerly separable features into new entities.

Downloads: 0