Introduction
In the classic rod-and-frame paradigm introduced by Witkin and Asch (
1948), a vertical rod inside a frame tilted in the fronto-parallel plane (roll direction) is seen as tilted in the opposite direction to the frame, implying that the subjective vertical has shifted in the same direction as the frame. This ‘rod-and-frame’ illusion (RFI) has received extensive investigation. There are large individual differences in the effect (Spinelli et al.
1995). Some subjects, termed ‘field dependent’ by Witkin and Asch, show large effects of the frame, while others show only small effects, or none at all. Other experiments, designed to test the influential perception–action model (Goodale and Milner
1992; Goodale et al.
1991), have sought to determine whether tilted visual reference frames affect action in the same way as they affect perception. At first sight, the results of these experiments have been contradictory. Dyde and Milner (
2002) reported that a tilted reference influenced perception (by the method of adjustment) but not action (distance between thumb and forefinger when grasping the rods). However, Craje et al. (
2008) found that the grip orientation used to grasp a rod was affected by a surrounding frame in the same way as an actual tilt of the rod from the vertical. Craje et al. argued that differences in task constraints may have contributed to the different findings (Smeets et al
2002). In the Dyde and Milner study, participants grasped the ends of the rod between their thumb and forefinger and were therefore dependent on the visual information regarding the
positions of the ends of the rod, while in the Craje et al. study (so these authors argue), orientation of the rod was the key information. However, Dyde and Milner (
2002) also report a perception–action dissociation for the RFI when the action was posting a card against a grating set in the center of a large tilted frame and conclude that the dissociation is
not specific to grasping. Therefore, the circumstances in which the RFI affects behavior are not at present clear.
The effect of a visual reference frame has also been investigated with the Roelofs effect, in which the apparent position of a small central target is altered when it is presented inside a large frame, out of left–right alignment with the observer’s median plane. Contrary to its effect of perception, the frame has been reported as having no effect on manual pointing or saccades (review by Cardoso-Leite and Gorea
2010). However, the perceptual task in this case consisted of comparing the perceived target position with that of memorized targets previously presented at various positions relative to the median plane. As Cardoso-Leite and Gorea point out, a parsimonious explanation for the apparent dissociation is that the frame causes a shift in the perceived position of the subject’s median plane. This would shift the perceived position of a target previously encoded relative to the median plane, but would not affect an eye movement planned in egocentric coordinates. This example shows the importance of clear reasoning before taking a perceptual-motor dissociation as evidence for ‘two systems,’ an imperative we were aware of in designing our experiments.
The purpose of the experiments we report here was to investigate the effect of visual reference frames on saccadic eye movements. Specifically, we wished to determine whether the control mechanism for saccades made in egocentric coordinates is affected by tilted frames of reference in the same way that apparent spatial relationships are. To examine this question, we used a task in which participants were required to make a saccade to a virtual target, rather than explicit target. We call this the ‘saccade-to-vertical’ task (Barnett-Cowan and Harris
2008). Specifically, observers fixated a point inside a circle and when the fixation point disappeared, attempted to make a saccade to a point on the circle vertically below the fixation point (the ‘6 o’clock’ position). There was no explicit visual target, other than the circle. The results we describe show that participants could perform this task accurately. In principle, planning a ‘saccade-to-vertical’ could rely on several sources of information. One might be the
perceived location of the 6 o’clock position provided by the visual context of the reference frame. Elsewhere, we have reported evidence that subjects can make accurate (oblique) saccades to a virtual target defined by a pointer presented in the context of a Poggendorff tilt illusion and that such eye movements are influenced by the visual context (Morgan and Melmoth
2011; Melmoth et al. in press). If visual information is mainly used to plan a vertical saccade, then the saccade should be affected by visual context in the same way as perception. Individual differences between ‘field-dependent’ and non-dependent subjects would also be reflected in their saccades-to-vertical. Another possibility is that the frame affects the perceived allocentric position of the target, but not the perceived egocentric position, in which case saccades should not be affected by the frame.
If not visual context, which sources of visual information do non-dependent subjects use? One possibility is that they are using a retinal coordinate system in which the vertical meridian is explicitly represented. Such a system would be vulnerable to head tilt and become less reliable than visual context. We predicted that head tilt would transform non-dependent individuals into field dependent. This was tested by combining the tilted and untilted frames with a 5.6° anti-clockwise head tilt, which is well beyond the range of compensatory cyclotorsion (Maxwell and Schor
2006).
To compare a frame’s effect on the saccade-to-vertical task with its effect on apparent spatial relationships, it was necessary to develop a perceptual test using the same stimulus. An obvious possibility would be to ask participants whether a briefly flashed dot was to the left or right of the 6 o’clock position (Tomassini et al.
2014). However, this ‘method of single stimuli’ is open to the critique that it confounds perceptual, decision and response biases (Morgan et al.
2012). A subject has only to respond in one of the two directions consistently when unsure of the correct answer to produce a bias in the psychometric function that is, in principle, indistinguishable from a genuine perceptual bias/illusion. This strategy need not necessarily be conscious. The reason for thinking that this may be important is that key results on context obtained with the method of single stimuli (Taya et al.
2009; Turi and Burr
2012) have not been confirmed using the two-alternative (2AFC) technique we describe here (Morgan
2013,
2014). The technique (technically ‘two-interval forced choice with a roving pedestal’) is to present two stimuli in temporal succession rather than one and to ask the subject to decide which of the two is
nearer to the 6 o’clock position. By varying the actual location of
both stimuli with a pedestal offset over a series of trials, it can be arranged that any genuine bias will affect the first and second stimulus equally and that the subject has no information on a given trial about which stimulus is actually closer to vertical. A decision bias of the kind ‘select the first stimulus if unsure’ will not mimic a genuine perceptual bias using this technique. A preliminary account of results obtained with this technique has been published elsewhere (Morgan et al.
2013).
In summary, we measured perceptual accuracy and context effects on the location of the apparent 6 o’clock position using a 2AFC method with a roving pedestal and saccade biases using ‘saccade-to-vertical.’ In both cases, three frames were used: (1) a control where the frame had zero tilt and two experimental conditions, in which the frame was tilted by (2) 5.6° in an anti-clockwise direction or (3) 5.6° in a clockwise direction. We also investigated the effect of a head tilt (5.6° anti-clockwise) under clockwise and anti-clockwise frame tilts, both for perceptual and for saccade tasks, with the hypothesis that tilting the participant’s head would make them more likely to use perceptual cues such as the frame and less likely to use gravity, thereby enhancing the frame effect.
Discussion
Our finding that a tilted frame of reference can displace vertical eye movements confirms that of Barnett-Cowan and Harris (
2008) but with considerably smaller tilts. Barnett-Cowan and Harris placed observers in a room that could be tilted relative to the vertical and asked participants to make repetitive vertical eye movements, either with respect to their head orientation or with respect to gravity. Both kinds of saccade were shifted from the vertical by tilting the room. In contrast to our procedure, there was no explicit target for the saccade, whereas we had a visible arc to which the participants were asked to make a saccade. A further difference is that we tilted only a single rectangular frame, rather than the whole room. Finally, our (5.6°) tilts were considerably smaller than those used by Barnett-Cowan and Harris (60° and 120°). It would be interesting to see whether the performances of our ‘frame-independent’ observers would be immune to an extremely tilted, immersive environment like this. If not, we would have to conclude that our ‘frame-independent’ observers merely gave a relatively low weight to our slightly tilted rectangle. In other words, the difference between them and ‘frame-dependent’ observers could be quantitative rather than qualitative.
Spering and Carrasco (
2015) have recently reviewed dissociations between visual perception and eye movements, for example, greater sensitivity of smooth pursuit than perception to threshold changes in target velocity (Tavassoli and Ringach
2010). Our findings do not show such a dissociation in the case of a tilted frame, which had similar effects on saccades and a perceptual task. Indeed, individual differences between participants were reflected in both their perceptual and motor responses. The only possible exception is the fact that EM showed no frame effect in the perceptual task (with head upright) and a small but consistent effect on saccades. This is just a hint that the saccade task may be more sensitive, but statistical reasons would need to be ruled out before this could be accepted as a real difference in relative sensitivity.
Previous investigations have revealed inconsistent effects of a tilted reference frame upon manual behavior, depending on the task (Dyde and Milner
2002; Craje et al.
2008). Our purpose was to see whether a tilted reference frame would affect eye movements in a ‘saccade-to-vertical’ task. It did. From this, we conclude that planning of a saccadic eye movement can be influenced by visual context, as also appears to be the case for the Poggendorff illusion (Morgan and Melmoth
2011; Melmoth et al. in press). We have also shown that individual differences, previously reported for the perceptual frame effect (Spinelli et al.
1995; Witkin and Asch
1948), are found in the saccade-to-vertical task too and that participants who have little or no perceptual bias have either no saccade bias or one in the ‘wrong’ direction. We also find that both perceptual and saccade biases due to a visual reference frame occur in a head-tilted condition, in which it might be anticipated that the subject’s egocentric encoding of gravitational vertical (i.e., straight down toward their feet) would be less affected by the visual context.
We initially thought it quite possible that saccadic eye movements might have a different balance of inputs from gravitational, visual context and body-frame cues from perception. The ability of subjects to make accurate second saccades in a double-step paradigm in the dark (Becker and Jurgens
1979) suggests that eye movements can be programmed in a body-centered coordinate frame. However, our results show that ‘frame-dependent’ subjects in the perceptual test are also ‘frame-dependent’ in the ‘saccade-to-vertical’ test, and regardless of their head position, arguing that they do not use a body—or other gravitational-centered reference frame for their saccades under either of these task conditions.