Top-down and bottom-up processes in vision

Abstract

To help us efficiently navigate our information-rich environment, the human brain has developed a set of cognitive mechanisms aimed at quickly identifying information that is likely to be behaviourally relevant, collectively referred to as selective attention. Attentional resources can be directed voluntarily, as occurs when we shift attention from one word to the next while reading; or involuntarily, such as when attention is captured by someone unexpectedly entering the room (Yantis, 2000). These two modes of attention are not as independent and clearly distinguishable as once thought (Gilbert & Li, 2013). For example, the extent to which a stimulus involuntarily captures attention is influenced by whether it possesses features relevant to an observer’s current task (the contingent capture hypothesis; Folk, Remington, & Johnston, 1992). Attentional capture is also affected by the level of working memory (WM) load an observer is under (Lavie, Hirst, Viding, & De Fockert, 2004), and by an observer’s WM capacity (Vogel, McCollough, & Machizawa, 2005). It has been suggested that maintaining top-down biases on attentional capture may require – and compete for – similar neural resources as WM (e.g., Lavie et al., 2004). If so, top-down influences on attentional capture should be reduced (a) when an observer is under high WM load, and (b) in individuals with a low WM capacity. In this thesis these possibilities were tested in a series of seven experiments in which human observers monitored visual stimulus streams for targets defined by a particular feature value (e.g., red). While observers performed these tasks, I took behavioural and electrophysiological measures of attentional capture by distracting visual stimuli that either possessed or did not possess the target feature value. First I replicated previous evidence for contingent capture. Distractors with the target feature value slowed behavioural responses to concurrently presented targets more than distractors with a non-target feature, consistent with the notion that distractors capture attention more strongly when they possess task-relevant features. Distractors with task-relevant features also evoked a larger N2pc component of the visual-evoked potential, an electrophysiological marker of the covert allocation of visual resources. I then investigated whether taxing WM reduces contingent capture. An n-back task that required information-updating eliminated the influence of task set on the distractor-evoked N2pc, suggesting that taxing WM compromised top-down biases on early visual processing. There was no such effect on a behavioural measure of attentional capture, however, possibly because taxing WM also compromised the transfer of information from early visual cortex to later stages of processing. A digit rehearsal task that required only information maintenance did not change the effect of task set on either the N2pc or behaviour. This pattern of results suggests that contingent capture is only reduced by WM load manipulations that place demands on the central executive component of WM (Baddeley & Hitch, 1974). Next I tested whether high WM load can reduce the effect of task set on even earlier stages of processing than the N2pc, specifically by measuring the P1 component of the visual-evoked potential. Unexpectedly, I failed to replicate a previous finding of an effect of feature-based task set on the P1. Because I found no effect of task set on the P1, I was unable to test whether such an effect was reduced under high WM load. In a final experiment I examined whether individual differences in WM capacity also influence contingent capture. Previously, Fukuda and Vogel (2011) found that individual differences in visual short-term memory (VSTM) capacity bore no relationship to contingent capture (although such differences did correlate with subsequent attentional disengagement). I set out to extend on this research by testing whether the capacity of the central executive component of WM influences contingent capture. Despite having multiple reliable measures of WM capacity and multiple behavioural and electrophysiological indices of attentional capture, there was no evidence for any relationship between WM capacity and contingent capture. In summary, I found support for the contingent capture hypothesis that distractors capture attention more strongly if they possess features relevant to an observer’s current task. I found that the task-dependent effect on early stages of processing was reduced when observers were under high WM load, provided this involved information-updating. This suggests that feature-selective mechanisms underlying contingent capture compete for shared neural resources with the central executive component of WM. This reduction in contingent capture did not affect later stages of processing, however, which suggests that taxing WM also compromises the transfer of information from early visual cortex to later stages of processing. Interestingly, I found no evidence for a relationship between contingent capture and measures of WM capacity, suggesting that the competition for resources between WM and feature-based task set only arise when both tasks are performed simultaneously.