Abstract
Proactive attentional templates shape our expectations and facilitate perception. Recent evidence has shown that templates are not necessarily veridical, but can be ‘warped’ within a feature dimension to facilitate optimal task performance. It remains unclear if these templates are additionally biased between feature dimensions – and whether templates become similarly biased when retrieved through long-term memory associations. Participants learned association between four shapes and four colored gratings. Gratings each had a unique combination of color (green or pink) and orientation (left or right tilt). On each trial, observers saw a shape followed by a grating and indicated whether or not the pair matched the learned shape-grating association. We manipulated the probability of the lure (non-match) stimuli either block-wise (experiment 1) or within a trial (experiment 2). In experiment 1, in some blocks the lure was most likely to differ from the target in color but not orientation, while in other blocks this was reversed. Participants were more likely to commit false alarms and to respond more slowly to unexpected lures, indicating that the template for the upcoming stimulus had been adapted such that the distinguishing feature dimension dominated the template. In experiment 2, we asked whether the same templates could be flexibly adjusted within a trial. This time, gratings either appeared after a short (1.25 s) or long (2.5 s) delay. If a lure appeared early, it would likely share one specific feature with the target but differ in the other. If a lure appeared late, the opposite was true. Here, observers showed higher false alarms to the unexpected lures from the short time period, irrespective of the time they were probed. The template was thus biased to the initially anticipated diagnostic feature, but this feature weighting was “sticky” and was not dynamically reversed within a trial.