Getting beyond the serial/parallel debate in visual search: a hybrid approach

The question of whether visual search involves at least one item-by-item serial processing stage or whether instead it is an entirely parallel process has been debated for decades, Recently, estimates of 'attentional dwell-time', which is the time required to reallocate atcention from one item to another, have been brought to bear on this question. Herc. we review this and other classes of evidence that favor serial or parallel models of visual search, and conclude thnt hybrid models that are neither strictly serial nor strictly parallel are better candidates for describing human visunl search. \Ve end the chapter with a sketch of one such model, and some of its possibilities. If you look for a 2 among the 5s in Fig. 9.1, it will take you longer to find the 2 in lB than in lA and even longer to confirm that there is no 2 present in IC. How should we understand the dependence of response time (RT) in this sort of task on the number of items (set size)? The answer to this question has been phrased in essentiaJly dichotomous terms for a generation. It could be that each item is processed one after the other in series (cf. e.g. Treisman and Gelade 1980; Wolfe et al. 1989; Wolfe 1994). It could be that information is accumulated from multiple items at the same time in parallel (Grossberg et al. 1994; Humphreys and Muller 1993; Palmer and rvicLean 1995). Which is it? Is attention deployed to one item at a time in visual search or is it distributed across many or all items? These sound like dramatically different alternatives. Why then has it proved so difficult for proponents of serial models or proponents of parallel models to gain the upper hand in this debate? One explanation has been that, for all their apparent differences, serial and parallel models can be made to predict similar experimental outcomes (e.g. Townsend 1971, 1976, 1990). We will argue for a somewhat different position. There is a class of plausible models