Attentional Synchrony

According to Dmytryk, the main trick of using saccadic eye movements to hide cuts is to get all viewers to move their eyes at the same time (Dmytryk, 1986; pg 438). This seems like a difficult thing to achieve considering that the pattern of eye movements made whilst viewing static images varies between and within viewers depending on such factors as individual differences, task requirements, and familiarity with the image (see Figure 3-6 and Yarbus, 1967). In every image there will be regions that are more likely to be fixated by all viewers. Visual features such as human faces, unusual or out of place objects, objects presented in the centre of the image, areas of high detail or significance to the viewing task will all be fixated more frequently (Yarbus, 1967). However, all viewers are free to visit any region of the image whenever they choose. This lack of control would not provide Dmytryk with the synchrony of attention required to hide a cut.

94 Figure 3-6: Seven records of eye movements by the same subject. Each record lasted 3 minutes. 1) Free examination. Before subsequent recordings, the subject was asked to: 2) estimate the material circumstances of the family; 3) give the ages of the people; 4) surmise what the family had been doing before the arrival of the "unexpected visitor;" 5) remember the clothes worn by the people; 6) remember the position of the people and objects in the room; 7) estimate how long the "unexpected visitor" had been away from the family (from Yarbus, 1967).

Fortunately, moving images introduce the possibility of the image (or dynamic visual scene as it now is) capturing a viewer’s attention instead of waiting to be fixated. The use of attention capturing visual features in film has already been established (see section 3.2.1), what is now of interest is whether such attention capture occurs across cuts.

This question has never been explicitly tested in an empirical study but several studies have performed similar enough investigations that by close examination of their results evidence of attention capture and attentional synchrony across cuts can be found. Studies recording the eye movements of viewers whilst they watch feature films have found that there is very little difference between where viewers look at

95 any point during a film (May et al., 2003; Stelmach, Tam, & Hearty, 1991; Tosi et al., 1997). The degree of agreement between viewers increases as the motion depicted in a scene increases and decreases (i.e. viewers look where they want) when the scene is more static (Tosi et al., 1997). Most fixations occur within a small region in the centre of the screen (a quarter of the width of the screen; Tosi et al., 1997). This central tendency of fixations has been explained as due to the convention of framing most shots so that they contain one significant object located at the centre of the screen (May et al., 2003). This tendency to centrally compose shots has also been associated with a lower frequency of eye movements and the impression of a film lacking in “visual momentum” i.e. being unexciting and “cinematically dead” (Block, 2001; Hochberg & Brooks, 1978a; Treuting, 2004).

Figure 3-7: Screen shots with superimposed gaze position from 19 viewers (pink and yellow spots). The frames number is presented under each image (30fps= 33.3ms per frame). Left-most image is the frame before the cut. Right-most is the frame at which all gaze positions finish moving after the cut.

A small exploratory eye tracking study was performed to generate illustrations of this supposed attentional synchrony39. This study presented a short film (2 minutes in

39 _{This study is not reported in full as, due to time constraints, the study was limited in terms of the} length and variety of films used (only one film 2 minutes in length was used) and the number of different viewing conditions that could be tested. Two different viewing conditions were used: silent

96 length) of a conversation edited by a professional editor40 according to the continuity editing rules. Figure 3-7 shows two types of cuts: a cut across dialogue with matching eyelines (top) and cut to close-up with matching action (a head turn; bottom). Superimposed on to the frames are spots representing the gaze positions of 19 viewers41. The clear clustering of gaze positions around the actors’ eyes can be seen. This is exactly as is expected given previous eye tracking evidence (Gullberg & Holmqvist, 1999; Klin, Jones, Schultz, Volkmar, & Cohen, 2002b; Yarbus, 1967). It appears that the larger the camera-subject distance (see Appendix A) the greater the degree of between-subject variability of gaze location (compare frame 1184 to all others). This is probably due to the increasing number of centres-of-interest as previously indicated by Hochberg & Brooks (Hochberg & Brooks, 1978a). The key effect illustrated by Figure 3-7 is the attentional synchrony occurring across cuts. The first shot of each sequence (frames 1115 and 1184) depict the gaze position immediately prior to the cut. The next frame occurs immediately after the cut (frames 1116 and 1185). Notice how the gaze position remains the same. The final shots depict the gaze position once all subjects have finished performing their first saccade after the cut (1125 and 1195). Notice how all subjects move their eyes to the same position. Any variability of pre-cut gaze position (e.g. frame 1184) has been eliminated by cutting to a new shot with a single centre-of-interest. What cannot be seen from this pattern of eye movements is if the saccades begin in anticipation of the cut as would be required for the cut to be hidden by saccadic suppression. The

(pink spots Figure 3-7) and audio (yellow spots). There were no differences between these groups in terms of where they looked and when they initiated saccades relative to a cut. This was highly unexpected and has been attributed to the low number of cuts tested and the infrequency of dialogue “bleeding” across cuts. For anything to be concluded about eye movements with and without sound a larger study would be required.

40 _{Chris Learmonth, Cinesite UK}

41 _{A dedicated piece of software called Gazeatron was developed by the author of this thesis to allow} gaze position from multiple viewers to be displayed at the same time. Gazeatron was developed in the Shockwave format (Macromedia, 2004). The screen shots presented in Figure 3-7 were generated by Gazeatron. The gaze positions are generated in real-time as the film plays which enables the experimenter to get a true insight into attentional synchrony. This function is not currently available in commercially available eye tracking software.

97 saccades, as illustrated here, take 300ms (top) and 267ms (bottom), which suggests that they were initiated in response to the cut. A larger, more controlled study is needed to look for anticipatory saccades as well as investigate the true pattern of attentional synchrony across cuts.

Luckily a study already exists that provides supporting evidence of anticipatory saccades across cuts (May et al., 2003). This study eye tracked viewers whilst they watched a popular feature film (The Mask of Zorro; Campbell, 1998) and then divided the data up into nine different types of cut. The length of saccadic eye movements occurring immediately after each cut was then calculated to see if the pattern of eye movements varied across cut types. May et al’s results indicated that there was an increase in the length of saccadic eye movements between 120 and 440ms after the majority of cuts (~77%). This indicated that the centres of interest differed across most shots. For some cuts, saccades were initiated in response to the cut (cuts such as those depicting the outcome of an action initiated in the first shot). Therefore, the cut could not have been hidden by saccadic suppression. However, for about 75% of cuts there were signs that some saccades were initiated prior to the cuts42. These types of cuts included cuts during conversations, cuts to over-the- shoulder shots, point-of-view shots, and close-ups. In general, all cuts where the object of the new shot was already present in the previous shot or had been cued in someway.

The type of cut showing the clearest signs of saccades coincident with cuts were those occurring during conversations. The continuity editing rules (specifically the 180° Rule) specify that such cuts must maintain the location of each character on the screen across the cut. The benefit of preserving screen location has previously been seen in relation to the pushing of attention across the cut by gaze shifts (see section 3.2.2). Now with May et al’s eye tracking results clear evidence of this benefit is seen. The social cues of the conversation, such as gaze shifts, head turns, speech and 42 _{The peak in saccadic eye movements started 120ms after a cut. Given that most saccades take} between 150 and 200ms to initiate and carry out (Goldberg et al., 1991), any saccade occurring less than 200ms after a cut is considered anticipatory.

98 hand gestures, direct the viewers gaze to a different character. The viewer initiates a saccadic eye movement and the editor cuts to a new shot of the same character at exactly the same time. By the time the viewer has recovered from their saccade 200ms later the character they saccaded to is still located in the same position but the shot has changed without them noticing. This suggest that editors construct their cuts to take advantage of viewers’ overt shifts of attention as proposed by Dmytryk (Dmytryk, 1986).