Visual Attention Glenn Mason-Riseborough (9/5/1997) Introduction The two experiments discussed in this essay relate to attention, and specifically visual attention. These experiments were conducted on students taking psychology paper 461.220 at the University of Auckland. The first experiment studies the difference between voluntary and reflexive orienting, while the second looks at the difference between serial and parallel processes in visual search. Visual Orienting Visual orienting refers to the focus of attention of the visual system and the way in which the eyes move to take in the surrounding environment. A distinction is being made between voluntary and reflexive orienting and one of the aims of this experiment is to compare reaction times between the two types of orienting. As the name suggests, voluntary orienting is the conscious orienting an organism does to change their attentional focus. On the other hand, reflexive orienting involves no conscious thought, and is an automatic reaction to changes in the environment. In this experiment the subjects were asked to perform two out of three tasks. The first task is intended as a voluntary orienting task and the other two as reflexive. Half the subjects did one reflexive task and the other half did the other task. The Tasks All three tasks were performed on a computer which calculated reaction times from the presentation of the target to the press of a key. The subjects were asked to focus on a central marker throughout the task. There were two outline boxes on either side of the central marker and the target could appear in either one of them. A cue would briefly appear before the presentation of the target. A factor that was studied over all three tasks was the delay between the onset of the cue and the onset of the target. This delay is referred to as the stimulus onset asynchrony (SOA). It is possible to detect the speed of the response to the cue, and attention orienting by varying the SOA. Another variable that was introduced over all tasks was the possibility of a misleading cue. For this, the cue would point to one of the boxes but the target would appear in the other. These were termed invalid trials and they occurred approximately 20% of the time, conversely valid trials occurred about 80% of the time. When the experiment was run, the central marker flashed briefly then the cue was presented for 66 milliseconds before the target appeared. The target was presented either 100 ms (short SOA) or 600 ms (long SOA) after the presentation of the cue. The three tasks differed in the way the cue was presented. The assumption was that a central cue produced voluntary orienting and a peripheral cue produced reflexive orienting. Central Arrow Cue: In this task, the cue consisted of an arrow pointing either left or right replacing the central marker. This task was intended as the voluntary orienting task. Brightness Peripheral Cue: For this task, the cue was a brightening at the outer corners of either the left or right box. Colour Peripheral Cue: The cue of this task consisted of the appearance of one or other of the outline boxes. The screen background was yellow and the boxes which appeared were a light shade of grey (as opposed to black and white respectively in the other tasks). These colours were chosen to minimise the apparent change of brightness (1). The target was black (as opposed to white). Results Results for the central arrow cue task were based on the performance of 163 subjects. The peripheral brightness cue task had 86 subjects, and the peripheral colour cue task had 79 subjects. The results are summarised in Figures 1, 2 and 3. The data for each of the tasks was analysed and two-tailed t tests were performed to test significance. For the central arrow cue, the reaction time for subjects was significantly faster for valid trials than for invalid trials. This significance occurred for both short and long SOA, with p<0.01 for both. There was similar results for the peripheral brightness cue in terms of significance. For both SOAs the valid trial was again significantly faster than invalid trials. The p values were again less than 1%, indicating strong evidence. The peripheral colour cues task provided slightly different results. For a short SOA (100 ms) the valid trials were significantly (p<0.01) faster than the invalid trials. However for the long SOA (600 ms) the invalid trials were significantly (p<0.001) faster than the valid trials. Analysis We can see from looking at Figure 1 that performance on the central arrow cue task uses voluntary processing. We can deduce this by seeing that the difference in response times is significantly larger between valid and invalid trials for the long (600 ms) SOA than the short SOA (100 ms) (the lines are diverging). This by further rationalised by the assumption that the subject takes a large amount of time (between 100 and 600 ms) to respond to the cue. For the short SOA the subject has not had time to process the cue, so valid and invalid trials take similar reaction times. For the long SOA the subject has processed the cue and consciously diverts his/her attention to the valid side If the target appears there, the reaction time is fast, if the target appears on the other side, reaction time is much slower. This diverging of lines does not occur in the Peripheral Brightness Cue task (Figure 2) or the Peripheral Colour Cue task (Figure 3), so we can assume that automatic processes are used for both of these. In fact, for the Peripheral Colour Cue task the lines converge and intersect. This can be understood by realising that the cue has been processed so quickly that attention has been diverted to the valid side and waited there so long that it has given up (so to speak) and returned to the centre or even the other side. There is a tendency not to repeat a search in the same area so when the target finally arrived on the valid side (after 600 ms), the mind is reluctant to focus attention on that side and thus the reaction takes longer. One reason that this did not occur for the Peripheral Brightness Cue task was that many subjects noted that the brightness cue was extremely difficult to see, so there may not have been as much reaction to it. Possible additional research would be to examine how the central arrow cues were processed. Perhaps if more neutral cues (such as L and R) were used, we may see different reaction times to these indicators. Peripheral cues could also be modified -- different levels of brightness or hue. Further research could also investigate different SOA times to see exactly what the reaction times are and exactly when the lines intersect for the Peripheral Colour Cue task. Visual Search In this second experiment, the subjects were asked to perform one out of two visual search tasks. The distinction being made in these tasks was also between types of automatic and controlled processing, however in this experiment the types of processing was in relation to Anne Treisman’s Feature Integration Theory (FIT). According to Treisman’s FIT, there are two types of perceptual processes that occur when we are searching for something (Treisman, 1985). The first is a preattentive level of processing. It is assumed that this type of processing occurs ‘automatically and across the visual field’ (Treisman, 1985, p 109). It is this process that is able to rapidly distinguish texture boundaries and it is used if the search is for something which can be distinguished by a simple feature. The second perceptual process requires attention and is thus slower. This operates in a serial manner and is used to search for objects with a number of different variables. It seems that each object must be examined in turn, for example, the task may be to find a green X in a square containing many different letters of many different colours, but the subject’s attention will have to focus on each letter in turn. The Tasks As with the visual orienting tasks, the visual search tasks were performed on a computer. For both tasks, the computer displayed a box which contained 1, 2, 4, 8, or 16 letters (the number of letters was chosen at random for each trial). The subject had to decide whether a target object (in this case a green X) was present or absent in the box and press a key corresponding to present or absent. The computer recorded the time from the presentation of the objects to the press of the key. Feature Condition: This task was intended as the parallel, preattentive visual search task. For this task there were two possible conditions chosen at random. Either the green X was displayed with blue and red X distractors, or the green X was displayed with green T and O distractors. In the former option the objects differed in terms of colour feature, in the latter, they different in terms of shape feature. Conjunction Condition: In this task the distractors differed in both colour and shape feature. The target (green X) was presented with green T and O distractors and also red and blue X distractors. This task was intended as the serial search task. Results For the feature condition, data was used for only the colour feature (the shape feature data was ignored). Results were based on the performance of 80 subjects for the Colour Feature condition and 74 subjects for the Conjunction condition. These results are summarised in Figures 4 and 5. Analysis Our data obtained in this experiment shows similar results to Treisman’s findings. Treisman (1985, p 109) states: We find that if a target differs from the distractors in some simple property, such as orientation or colour or curvature, the target is detected about equally fast in an array of 30 items and in an array of three items. If we were to show this result graphically we would find that we would get an approximately straight horizontal line across the display sizes. Figure 4 shows this trend for the present trials of the Colour Feature condition. The average time taken for all trials for each of the display sizes was approximately 600 ms and there is only a very small positive gradient. This indeed shows parallel processing, since the features of the objects must all be processed at the same time in order for there to be no increase in time with an increase of objects. The absent trials of the Colour Feature condition show similar results except for a small upward spike for small display sizes. For all other display sizes the mean response time is approximately 700 ms. This upward spike can be accounted for by the experimental set-up used. The colours chosen for the distractor objects was red and blue. On a black background these colours were extremely difficult to see (green could be seen without difficulty). If there were few letters present (without the green X) then it took some time before the subjects realised that the letters had indeed appeared on the screen. Consequently the time was not used for processing the features but for detecting the next trial. This problem could be rectified by using different colour combinations, for example white, yellow, or maybe orange, or alternatively using a different colour background (for example yellow). Our results for the other task performed (Conjunction Search condition) also show similarities to Treisman’s own findings. Treisman (1985, p 109) states: ... we find that if a target is characterised only by a conjunction of properties ... or it is defined only by its particular combination of components ..., the time taken to find the target or to decide that the target is not present increases linearly with the number of distractors. ... As distractors are added to the displays, the search time in positive trials therefore increases at half the rate of the search time in negative trials. It can be seen from the graph in Figure 5 that in general our results follow these same trends. As with the Colour Feature search we have an aberration, with a small spike for small display sizes of absent trials. This can again be explained by the perception difficulties of the colour when there was no green X. Treisman notes that the time taken to respond increases linearly as the display size increases. This can be seen in Figure 5 by the approximately constant positive gradient of both lines across the display sizes. Treisman also notes that present (positive) trials increase at half the rate of absent (negative) trials. This is slightly more difficult to see in our graph (Figure 5), however it is clear that the present trials have a smaller gradient than the present trials and the two lines are definitely diverging. As explained above, the experimental design could be improved by choosing more appropriate colours for the objects. Also there could be a greater number of different display sizes. From looking at the graphs (Figures 4 and 5), it can be seen that there are large parts (namely between 8 and 16 objects) that we do not know anything about. Also it should be noted that we do not know if the trends continue beyond a display size of 16 objects. On the other hand, the number of display sizes should not be so large that the task becomes too long and the subjects become fatigued. As Treisman (1985) states, there are practical uses for this type of research. Further study may be useful for industry in quality control, advertising, or warning signs (for example road signs). Conclusions These experiments (visual orienting and visual search) show the difference between automatic and controlled mental processes. From both experiments it can be seen that automatic processes are faster, but not subject to conscious control or awareness. We can see that the automatic processes were able to be performed in parallel and large numbers of objects could be processed with little increase in time required. However, this was only able to occur for simple feature differences. Controlled mental processes were able to process relatively complex objects with multiple components but this could only occur in serial. Large numbers of objects took longer to process than small numbers of objects. References: Lambert, T. (1997). Unpublished laboratory notes for University of Auckland paper 461.220. Treisman A. (1985). Features and Objects in Visual Processing. Scientific American, 254, 106 - 115. Footnotes: 1 Patrick Cavanagh of Harvard University (cited in Lambert (1997) advocates using a white cue on a yellow background to minimise the apparent change in brightness.