The role of priming in conjunctive visual search

Abstract

To assess the role of priming in conjunctive visual search tasks, we systematically varied the consistency of the target and distractor identity between different conditions. Search was fastest in the standard conjunctive search paradigm where identities remained constant. Search was slowest when potential target identity varied predictably for each successive trial (the ‘switch’ condition). The role of priming was also demonstrated on a trial-by-trial basis in a ‘streak’ condition where target and distractor identity was unpredictable yet was consistent within streaks. When the target to be found was the same for a few trials in a row, search performance became similar to that when the potential target was the same on all trials. A similar pattern was found for the target absent trials, suggesting that priming is based on the whole search array rather than just the target in each case. Further analysis indicated that the effects of priming are sufficiently strong to account for the advantage seen for the conjunctive search task. We conclude that the role of priming in visual search is underestimated in current theories of visual search and that differences in search times often attributed to top-down guidance may instead reflect the benefits of priming.

Introduction

The importance of priming in various cognitive tasks is well known and has been researched extensively. Its role in visual search, however, has been poorly understood and relatively neglected. In a typical visual search task observers search for the same target repeatedly among a set of distractors. Yet, in some variants of the task, the identity of the target and distractors can change from one trial to the next. Although the effects of target variation have been demonstrated (Maljkovic and Nakayama, 1994, Maljkovic and Nakayama, 1996, Maljkovic and Nakayama, 2000, Schneider and Shiffrin, 1977) current theories of visual search do not address the issue of repeating targets or distractors (Duncan and Humphreys, 1989, Grossberg et al., 1994, Palmer, 1995, Treisman and Gelade, 1980, Wolfe, 1994).

The characteristics of repetition effects in visual search have been documented by Maljkovic and Nakayama, 1994, Maljkovic and Nakayama, 1996. They found short-term priming for features and locations of the target in a feature (“pop-out”) search task. Thus, if the target had the same features, or appeared in the same location as on the previous trial, search was faster than otherwise. These priming effects were very strong, effective over a number of trials, and not subject to voluntary control or expectation.

Why should priming of this sort be important for theories of visual search? Firstly, the empirical basis of many such theories is experiments in which the same target is to be found repeatedly among the same distractors. Secondly, priming effects may be large enough to contaminate the theories themselves, for example when differences in search rates between experiments where the target varies and where it remains the same are not attributed to this difference in target identity, but are instead spuriously attributed to some other difference between the two experiments. Thirdly, organisms do not normally spend their time scanning the visual environment for completely novel objects from one moment to the next but instead the targets tend to be relatively stable over time. In fact, when engaged in a simple task requiring input from the visual system we seem to make surprisingly many eye movements to the same targets. In one informative experiment observers' eye movements were monitored while they were engaged in copying a pattern of colored chips (Ballard, Hayhoe, Li, & Whitehead, 1992). Their subjects did not seem to memorize the layout of the chips they were to copy but made frequent eye movements back and forth between the copying pattern and the area where the copied pattern was to be placed. Maljkovic and Nakayama (1994) suggested that the priming of pop-out played a role in this process by allowing fast return of gaze to previously inspected areas or features. Similarly, in the wild, a predator tracking the movement of potential prey would benefit greatly from being able to reorient quickly to the tracked prey.

Given the strength of priming and its neglect in theoretical accounts of visual search, a study of its role is undertaken here. Our goals were three-fold: first, to assess the degree to which performance in visual search can be influenced by and accounted for by priming; second, to assess the nature of priming itself; and third, to investigate what role priming should play in theoretical accounts of visual search.

It is useful before proceeding further to delineate what we mean by the term priming. We think of priming as constituting an altered representational state for a feature that results in more efficient processing of this feature than otherwise. Thus, in visual search experiments a primed feature associated with a target will be found faster than others. Similarly, it is conceivable that primed distractors can be more easily rejected on subsequent occasions as the research on “negative” priming (see Tipper, 1985, Tipper, 1992) suggests. The priming pattern should then be observable from differences in reaction times to repeated vs. unrepeated features. Furthermore, priming effects can last for some time, but the largest priming effects seen in visual search experiments are relatively short lived, extending over thirty seconds or less (Maljkovic & Nakayama, 1994).

Priming as we have defined it differs significantly from putative top-down processing. First, it does not require conscious effort or explicit knowledge. Second, it is impervious to prior knowledge or expectation (Maljkovic & Nakayama, 1994). Priming is also distinct from perceptual learning. This latter phenomenon is retinotopic, or confined to particular locations (see e.g. Dosher and Lu, 1999, Goldstone, 1998, Karni and Sagi, 1993), whereas priming operates in object-based rather than retinotopic coordinates (Maljkovic & Nakayama, 1996; see also Kristjánsson et al., 2001, Kristjánsson and Nakayama, 2002). Interestingly, Sireteanu and Rettenbach (2000) showed that learning in visual search tasks affects structures at relatively high levels of the processing stream, since the learning they observed generalized between visual fields, over different locations and between the two eyes of their subjects. Their effects, however, were observed over months of training, so the relevance to the present studies is at present unknown, since our following study addresses between-trial priming in a visual search task.

To investigate the role of priming in visual search, we compared performance in several variants of a conjunctive visual search task. In conjunctive search, there is no single feature property that defines the target. The identity of the target is determined by the conjunction of two features (Treisman & Gelade, 1980). In our baseline condition, which we call conjunctive search, the target and distractor identities were stated to the observer and were consistent over a whole trial block. The features we used are horizontal vs. vertical and red vs. green. Thus, a target could, for example, be a red vertical bar among horizontal red and vertical green distractor bars. Three variants of conjunctive search were also used, designed to reveal the presence and characteristics of possible priming effects. (1) In the “switch” condition the potential target always changed orientation from one trial to the next. (2) In the “streak” condition the potential target retained its orientation for longer stretches of adjacent trials. (3) In the “random” condition, the potential target orientation changed randomly from one trial to the next.

The target was always a red horizontal or vertical bar. As in all conjunction tasks, the orientation of the target defined the distractors. Thus, when the target was horizontal, the red distractors were vertical, and the green distractors were horizontal. When the target was vertical, the distractors were red horizontal and green vertical bars. Fig. 1 presents a typical search array and Fig. 2 shows a representative sequence of the target identity on ten consecutive trials under the conjunction, switch and streak conditions.

Our operative assumption is that priming plays a significant and perhaps decisive role in conjunctive visual search. We can thus make a number of predictions as to the relative efficiency of visual search in the various conditions outlined. In the baseline conjunction condition the target remains the same on every trial. As such, it will most fully benefit from priming and consequently should yield the fastest search times overall. In contrast, performance should be worst under the “switch” condition since the same target never repeats and there is little opportunity for priming to develop. Priming benefits are likely to be larger in the streak than switch condition, and overall search performance should then be better under the streak than the switch condition but not as good as under the conjunction condition since priming there should be maximal, because the target is always the same.

In the streak condition, the orientation of the target repeats for long stretches of continuous trials (“streaks”). Performance should then be better in the later part of a streak, compared to the beginning of a streak, since in the later part, the target has been the same for a number of the preceding trials. In other words, priming should build up over a streak with consequent reductions in reaction times for later positions in a streak. The target was only present on half of the trials. We made no a priori assumptions that priming occurs only for the target. It is certainly possible that priming could also take place for the orientation of the distractors, thus leading to their faster rejection as non-targets. In fact, Maljkovic and Nakayama (1994, Experiment 8) showed that their priming effects on response times were both due to target facilitation and distractor inhibition.