Differences between visual and memory search: Implications for models of attention.

摘要

Automatism and attention involved in visual search for target letters is investigated. A number of theories are examined, including Fisher's Feature Overlap Model (1986; Fisher & Duffy, 1988), Treisman's Feature Integration Theory (e.g. Treisman & Gelade, 1980), Duncan & Humphrey's Resemblance Theory (1988) and Shiffrin & Schneider's Automatism Theory (1977; Schneider & Shiffrin, 1977). Previous research has suggested the importance of 2 training paradigms: varied mapping (VM) and consistent mapping (CM). In consistent, but not varied mapping, automatic target detection appears to develop. However, previous research did not fully match CM and VM training conditions. Therefore, in all the experiments described, the amount of training on particular target-display combinations was equated between consistent and varied mapping training conditions. In Experiments One, Two and Three, when memory load was small, no consistent mapping advantage was seen. This may have been the result of subjects choosing to use optimal feature comparison sequences in both VM and CM conditions, when the load was small enough to permit such a strategy. However, when memory load was high, subjects came to rely on automatic search systems in consistent mapping, but not varied mapping conditions. The last experiment was carried out using a pure memory search task . A clear CM advantage in search emerged within the first session, even though the amount of training on CM and VM memory set and target combinations was equated. Also, this CM search advantage was observed using the same kinds of stimulus items which had not produced a CM advantage during pure visual search. The results of the four experiments are described within the framework of a hierarchical search model, in which stimuli are coded (using attention) to various levels, from featural to categorical. When responses can be based on a strong association to the categorical level, due to consistent training, search can be carried out in automatically. At the same time, efficient search can be based on comparisons made only at the featural level, when the load is small enough during search to allow such a strategy. When large loads or inconsistent training prevent responses based on one of these two levels during search, a limited capacity feature search must be carried out, often requiring serial comparisons through the display.