|
Article Excerpt
Recently, many vision scientists have been fascinated by the striking difficulty people have detecting visual changes that occur simultaneously with some kind of disruption. The phenomenon is known as change blindness (CB) and can be induced when the transient motion signal the visual system normally relies upon to detect changes in the visual array is masked (see Rensink, 2002; Simons, 2000; Simons & Levin, 1997). Changes between successive views of a scene often remain undetected if the change coincides with a saccade (e.g. Hollingworth & Henderson, 2002) a blank screen or flicker (e.g. Rensink, O'Regan, & Clark, 1997), multiple motion transients or 'mud-splashes' (O'Regan, Rensink, & Clark, 1999), a film cut (e.g. Levin & Simons, 1997) or momentary occlusion of real-world objects (e.g. Simons & Levin, 1998). These failures of change detection are often quite dramatic and have sparked lively debates about the nature of visual representation and memory. Does CB indicate that our visual representation of the world is exceedingly impoverished, as some theorists have suggested (e.g. Rensink, 2000, 2002; Simons & Levin, 1997)? Or might CB occur even though our visual representations are relatively rich and detailed, as others have suggested (e.g. Henderson & Hollingworth, 2003a; Hollingworth & Henderson, 2002; Hollingworth, Williams, & Henderson, 2001)?
Although these two explanations might seem contradictory, we argue that they are best integrated, and that combining the impoverished and rich representation views leads to the prediction that it is possible to have situations yielding poor change-detection performance but good long-term visual recognition performance. This prediction is particularly counter-intuitive because CB would appear to represent an encoding failure, and therefore a barrier to long-term recognition. Thus, the long delay between initial viewing and test that characterizes long-term recognition should work against long-term recognition performance being better than change detection. In spite of this, several recent studies have demonstrated that people can retain visual information in longer term memory (e.g. Hollingworth & Henderson, 2002), and a few have demonstrated this even in the context of CB (e.g. Angelone, Levin, & Simons, 2003; Simons, Chabris, Schnur, & Levin, 2002). However, several limitations on this work have made it difficult to demonstrate good memory in the face of CB convincingly.
Change blindness and visual memory
As mentioned above, CB is often taken to suggest that people represent very little visual information. However, this is not necessarily the case, and a number of studies have demonstrated that people do often see relatively subtle visual changes and that they are able to remember visual information from scenes. For example, Hollingworth and Henderson (2002; Henderson & Hollingworth, 2003b; Hollingworth, Schrock, & Henderson, 2001) used a saccade contingent change paradigm to demonstrate that token-level changes (i.e. substituting one object with a different object from the same basic-level category; e.g. one guitar with another guitar) and object rotations could be detected at above-floor levels during on-line viewing of a scene. Since neither type of change drastically alters the conceptual or semantic nature of the changed object or scene, the representations required to detect these changes are presumably visual. In addition, Hollingworth and Henderson (2002; Castelhano & Henderson, 2005; Hollingworth, 2004; Hollingworth et al., 2001) also found that participants were able to accurately distinguish novel and previously viewed object tokens and orientations on a two-alternative forced-choice (2-AFC) test given several minutes after initial viewing of the scenes. These findings clearly demonstrate that people can create visual representations, retain them and use them to detect changes, and further provide evidence that change detection can be enhanced when observers are required to attend both to pre- and post-change objects. This is an important caveat to previous CB findings that may have occurred in part because participants were not required to do this (see also Hollingworth, Schrock, & Henderson, 2001).
However, it is important to distinguish between what people can do and what they generally do. Much of the research investigating visual memory involves intentional change-detection tasks in which participants are told they will have to detect changes. Therefore, any representational successes observed in these tasks could be specific to situations in which observers engage in purposeful task-specific storage, elaboration, and rehearsal (see Mack & Rock, 1998, for a review of the argument that intentional tasks invoke strategies that are often not present in more natural incidental tasks). Thus, these findings are more about people's representational capacity than about the degree to which visual representation is inherent to (or characteristic of) scene perception outside this context.
It is therefore useful to consider the degree to which people create and use visual memories when they are not required to. In many natural situations, various visual processes are necessary to understand a scene's meaning or to actively complete a specific task, but the creation and utilization of explicit visual memories may not be among them. In such cases, encoding of visual details is incidental to the task at hand, and we can ask whether visual representations are not only created, but also used as part and parcel of more general processes.
There is good evidence that people create rich visual memories in the absence of explicit task demands. For example, Castelhano and Henderson (2005) recently demonstrated that people are able to accurately distinguish previously viewed object tokens from same-category foils, even on surprise recognition memory tests. This result clearly shows that intention to study for a memory test per se is not necessary to succeed on a token-level long-term recognition test. However, these results do not address the question of whether such a rich visual memory is typically used to monitor the consistency of objects over time, because Castelhano and Henderson did not measure incidental change detection.
When observers are not looking for changes, the most basic finding is, unsurprisingly, that change detection is less frequent (e.g. Levin & Simons, 1997). However, the interesting question is whether, in these cases, it is possible to get evidence that detailed visual memories nonetheless exist. Such a finding is logically possible because, as a number of authors have pointed out, CB can occur when sufficient representations of pre- and post-change objects exist, but are simply not compared (Angelone et al., 2003; Hollingworth, 2003; Scott-Brown, Baker, & Orbach, 2000; Simons, 2000). For example, Angelone et al. asked observers to view videos in which an actor's clothes changed across a cut, or in which the actor was substituted for another actor across a cut. The change occurred while the actor was presumably the centre of attention (the cut occurred while she was speaking), and observers often failed to detect the change. More important, observers who missed the change were subsequently able to recognize the pre-change actor, and further, were no less accurate than observers who saw the change. Accordingly, it appears that visual representations are created and remembered in incidental tasks, and CB cannot only underestimate the prevalence of visual representations, but can be entirely uncorrelated with them in some cases (but not all; see Levin, Simons, Angelone, & Chabris, 2002).
However, these findings have a number of crucial shortcomings as evidence that long-term representations are created in the absence of task-specific elaboration but not used for the kind of monitoring and comparison processes that are inherent to change detection. First, Angelone et al. (2003) did not monitor what observers were attending to during the films. This is important because it is possible that people who missed the change did so because they elaborated on features of the pre-change actor but not the post-change actor. In contrast, participants who detected the change may have done so because they elaborated on features of both actors. Accordingly, recognition of the pre-change actor would have been equivalent between participants who saw the change and those who did not, because the recognition test assessed memory for the pre-change object. In addition, although this was an incidental change-detection task, participants may have been intentionally elaborating upon visual features simply because they were in the laboratory watching a film they expected to be asked about. Thus, while participants were not told to do a change-detection task, they were not given a replacement for it and were instead left to their own devices. In summary, these experiments clearly suggest that people sometimes create visual memories in incidental tasks, sometimes even when they have not detected changes. However, there remains no clear demonstration of successful visual memory in the context of unsuccessful change detection. It is the aim of the current set of experiments to provide such demonstrations.
Current experiments
To test whether it is possible to observe a global failure of change detection in the context of good visual memory, the current experiments take several precautions to create conditions under which participants are known to have focused attention on both the pre- and post-change objects while not being encouraged to intentionally elaborate on visual features. To this end, the current experiments adopt an incidental change-detection procedure, but unlike previous studies, participants are given one of two concrete cover tasks. In both tasks, participants view a scene in which locations are successively cued (see Fig. 1). For both tasks, the changing object is cued before and after the change, ensuring that observer's had a chance to encode relevant information. In the object task, participants respond when the cue appears around a location that contains an object, but not an empty location (Experiments 1, 2 and 3). In the upright task, participants respond when the cue appears around an object that is upright, and do not respond when the cued object is inverted (Experiments 1 and 2) or if the cued location is empty (Experiment 2). Notice that these tasks differ in that the object task can be performed based upon 'pre-attentive' representations, whereas the upright task probably requires attention. This conjecture is supported by visual search experiments, which typically find that searches for inverted objects are inefficient (Wolfe, 2001; Wolfe, Klempen, & Shulman, 1999).
[FIGURE 1 OMITTED]
An additional precaution was to use a forced-choice measure of change localization that is more sensitive than simply asking for an explicit report of change detection. Forced-choice localization tests have been used previously to demonstrate what might be called, in theory neutral terms, subtle visual memories in the face of CB (Fernandez-Duque & Thornton, 2000, 2003; but see Mitroff, Simons, & Franconeri, 2002). In previous experiments, however, the forced-choice displays contained visual information from the post-change display, and are therefore not useful as a way to measure incidental change detection. Basically, change localization tests give participants a second chance to detect a change, and this second chance must be intentional because participants must know why the forced-choice test is being administered. Thus, giving participants a localization test that contains visual information about the changed arrays gives them a chance to intentionally detect a change that was potentially missed when it was unexpected. Since the current experiments seek to measure incidental change detection, we did not want to give...
|
