Despite the evidence for poor explicit change-detection across saccades, recently evidence o f implicit change-detection has been suggested within a number of different paradigms. Hayhoe, Bensinger and Ballard (1998) have focused specifically on processing during ‘active vision’ (see also, Ballard, Hayhoe and Peltz, 1995; Hayhoe, 2000); that is, examining the functioning of saccades and fixations while subjects
interact with the environment in order to perform a particular task. The task Hayhoe et al have used to examine change-detection in active vision involves subjects copying a pattern of coloured blocks (see Figure 1.7). Subjects are required to copy, within the workspace area, the pattern displayed in the model by selecting blocks from the resource area. Model Workspace
A .
Before pick-up Resource area WorkspaceB .
After pick-up Resource areaFigure 1.7. Example of the two trial types from Hayhoe et al (1998). Subjects are required to copy the ‘model’ pattern with coloured blocks from the ‘resource a r e a ’, into the ‘workspace’ whilst their eye movements are analysed.
The two black arrows indicate the direction of the saccade when changes could be made to the model.
A - The saccade from the workspace to the model before selecting the next block. B - The saccade from the resource area to the model after picking up the next block.
Their work has revealed that explicit change-detection (for changes occurring in the model) during these tasks is surprisingly poor, despite the fact that the model is highly relevant to the task that subjects are completing. However, there is some evidence for implicit detection of change when subjects’ fixation times on the model
are examined. This measure revealed that subjects would fixate longer on the model area after a change, despite the failure in explicit detection. Furthermore, the length of this fixation extension was mediated by the purpose o f the particular saccade. For example, subjects fixated longer if the changed item has already been indicated as being useful to the present task. This was demonstrated by the fact that when a change was made to the model after pick-up from the resource area (see Figure 1.7B) there was a greater effect upon subjects’ fixation duration, as compared to changes occurring before pick-up (see Figure 1.7A). Hayhoe et al interpreted this in terms of the fact that the function of the saccade made prior to pick-up was to assess colour information in order to select the next block, therefore at that particular point none of the blocks are specifically relevant. However, after pick-up, the saccade back to the model is performed to assess the correct location to place a particular block, therefore any changes in block formation may now interfere with the representation of the pattern the subjects have when selecting the correct colour for that particular move. Hayhoe et al also manipulated the number of blocks changed and discovered that fixations on the model increased as the number of changes to the blocks increased.
Earlier pioneering research carried out by Castiello, Paulignan and Jeannerod (1991) compared explicit detection of change with motor performance related to this change. They demonstrated that subjects could immediately correct a hand movement if they were reaching for an item that was suddenly displaced. However explicit detections of this change apparently occurred only later, often 300 msecs after the motor corrections had been made. Thus, the authors suggested that awareness of the change occurs after the motor system has already processed and reacted to it (see also, Castiello and Jeannerod, 1991).
Hollingworth, Williams and Henderson (2001) also suggested that change detection was underestimated by previous methods for studying both ‘Change Blindness’ in a flicker paradigm, and also in the examinations of similar phenomena within saccadic research (see also, Hollingworth & Henderson, 2000; Hollingworth & Henderson, 2002). The task they utilised involved subjects scanning black-and-white lines drawings of natural scenes after being instructed that they would receive a memory test on them later. They were also told that changes could occur to the pictures while they examined them and that they should press a response key if this occurred. Results of Hollingworth et al suggested that some processing o f changes occurred without explicit detection. Their measure of this processing was similar to Hayhoe et al, as subjects’ fixation duration was seen to increase when their eyes returned to a changed item. This effect was often independent from any explicit detection of change. Hollingworth and Henderson (2000) reported a similar effect with full-colour 3D images. Further evidence of implicit change detection across saccades was suggested by other data from the same research group (Hollingworth, Schrock & Henderson, 2001). Within this study there was evidence that saccades were faster when saccading towards a changed item than a non-changed item, even in the absence of explicit change detection.
Over and above these implicit effects, Hollingworth, Williams and Henderson (2001) have argued that even explicit change-detection was actually better than reported by other experiments in transsaccadic and ‘Change Blindness’ research. They reported 27% overall explicit detection, which they claimed to be above the level that subjects would detect if many researchers were correct in suggesting that internal visual representations are extremely sparse (O’Regan, 1992; Rensink, O’Regan & Clark, 1997, Rensink, 2000; O’Regan & Noe”, 2001). Furthermore, they
noted that a significant number of detections occurred some time after the actual change (over 1500 msecs later) and that over 90% of these late detections were during re-fixations of the changed item. This fact appears suggestive of some relatively detailed preserved memory for the scene across fixations, Hollingworth et al stated that they believed that “detailed visual information was often retained for a relatively long period of time and consulted only when focal attention was directed back to the [change]”. This argument is in clear contrast to the proposal discussed earlier for very poor internal visual memory being supported by, for example, the ‘external memory’ provided by the outside world (O’Regan, 1992).
Furthermore, Henderson and Hollingworth (1999) have examined in detail conditions under which changes can be explicitly detected across saccades (see also, Hollingworth, Schrock & Henderson, 2001). They examined whether change- detection was affected by whether the change occurred in an item that was the current target of the saccade, the site of the previous fixation, or another item in the scene. Their results suggested that changes were detected better to items that were the target of the ongoing saccade. However detection of change (particularly deletions from the scene) were also good if they occurred to the previously fixated item (87% versus 78%). Occasionally, changes were subsequently detected on refixation, as in the study discussed above.
Thus, there is some evidence to suggest that those arguing for sparse visual representations and poor visual memory, both without attention and across saccadic eye movements, may have underestimated the extent of underlying processing, especially by failing to consider implicit measures sufficiently.