Abstract
Dynamic dot quartets (DDQs) appear to jump along the horizontal orvertical sides of an implicit square. The axis of motion is ambiguous and bistable, but an array of DDQs all appear to share a single axis of motion at any instant (Ramachandran and Anstis, 1983). To study the grouping of ambiguous objects, we exploited the fact that DDQs can be clamped into a definite perceptual state by manipulating the ratio of path lengths. In a typical experiment, all the DDQs in an array are clamped for several seconds, and then the aspect ratio is changed to one, and the post-clamp switching dynamics observed. Last year we reported that this procedure leads to an orthogonal axis aftereffect (OAE) with the following characteristics: 1. there tends to be a short latency post-clamp perceptual switch; 2. the OAE duration grows with increased adaptation; 3. the OAE is not retinally-specific; 4. Most observers cannot see simultaneous aftereffects following simultaneous orthogonal clamps. Together these imply that the OAE does not depend on adaptation of local motion detectors, but rather on an inferred global transformation. We now report that the OAE depends critically on attention. An attention-demanding task was added to the standard perceptual clamp experiment. Two seconds into the clamp phase a Landolt C appeared that switched among four orientations at 1 Hz and disappeared at the end of the clamp phase. Observers reported the direction of the C when it was visible, and the DDQ axis otherwise. In control trials observers ignored the C and reported only DDQ motion axis. In these trials the OAE peaked 7–10 seconds into the test phase and disappeared within 20 seconds. In contrast, in trials with the attention-demanding C task, the OAE was completely abolished in all observers. The attention-dependence of the OAE supports the evidence from our previous experiments that the OAE depends on active invocation of a transformation of a scene model.