Procedure and stimuli.

At the beginning of each trial, a word cue was presented at the center of the display (e.g., “Bear”), indicating the present trial’s search target. After a variable inter-stimulus interval (ISI), objects appeared for 1500 ms or until a response was made by the participant. The next trial started after a 1000-ms delay.

Results for Experiment 1.

A, Normalized response time was longer for horizontally oriented objects (90°) than for upright objects (0°) for set sizes 4 (purple line) and 9 (red line). B, The accuracy was higher when objects were upright than when they were horizontal. Format as in A. C, Normalized response time increased as a function of set size for upright (red line) and tilted objects (purple line). D, Difference in intercept and slope of the response time–set size function (lines shown in C) between upright and horizontal objects. Error bars represent S.E.M.

Illustrations of two possible mechanisms for orientation dependency and additional results for Experiment 1.

A, Top row, an internal reference frame is rotated to align with objects at the beginning of each trial. Bottom row, each object is mentally rotated to a familiar orientation during serial search. B, Top, reference frame transformation predicts an orientation dependency on the intercept of the RT-set size relationship, as the transformation step is independent of the number of objects. Bottom, object rotation predicts a difference in slope between upright and tilted objects, because each object requires additional time for mental rotation. C, Longer normalized response time was found for both target-absent (dashed lines) and present trials (solid lines) for set sizes 4 (dashed blue and solid red lines) and 9 (dashed purple and solid yellow lines). D, Response time varied as a function of effective set size for upright (red) and tilted objects (purple). Lines represent linear fit to the data. E, Difference in intercept and slope of the response time–effective set size functions shown in D. Error bars indicate S.E.M.

Apparatus in Experiment 2.

A, Participants sat in a flight simulator, which allowed full body roll on a block-by-block basis, separating the egocentric reference cues from visual context and gravitational cues. B, The virtual scene presented on the head-mounted display was rotated independently to manipulate visual context. C, The internal reference frame can be modeled as a vector sum of visual context, egocentric, and gravitational cues with unknown weights. While not directly manipulated, gravitational cues can be dissociated from the other two cues by rotating the visual context and the participant’s body in the same direction.

Experimental conditions in Experiment 2.

A, Baseline condition, in which visual context (represented by the tree and cloud), egocentric (the observer’s head and body), and gravitational (gray arrow, g) cues were all upright. 0° object orientation indicated upright with respect to all three cues. B, Visual context condition, in which only the visual context was rotated by 90°. In this case, 90° became upright with respect to the visual context, and 0° was upright with respect to the other two cues. C, Egocentric condition, in which the observer’s full body was rotated by 90°. Here, 90° was upright with respect to egocentric cues, while 0° remained upright to the other two cues. D, Gravitational condition, in which both visual context and the observer’s body were rotated by 90° in the same direction. In this condition, 0° was upright to visual context and egocentric cues, and 90° was upright with respect to gravity.

Results for Experiment 2.

A, Normalized response time in each experimental condition. 0° and 90° represent upright and tilted objects, respectively. In each condition, object orientation was defined relative to the reference cues not under experimental manipulation (i.e., gravitational reference frame in the Baseline, Visual context, and Egocentric conditions; egocentric reference frame in the Gravitational condition; see Figure 5). B, Weights for each reference frame inferred from changes in orientation dependency. Error bars represent S.E.M.