How do we use the past to predict the future in oculomotor search?

occulomotor_prediction

When we conduct visual search for an item of interest we can not only ask ‘what looks like my target’ but we can also ask ‘where do I expect the target to be?’ In a forthcoming paper in the journal Vision Research I asked: how do we use our past experience to construct predictions of where a target may be?

People conducted many repeated searches for a target that could be in 1 of 4 locations on each search trial. Each location is indicated by a white square, so there is no visual indication of where the target is. An eye tracker was used to only reveal whether a target was or was not found when one of these locations was directly fixated. By examining the order in which people searched each location, inferences could be made about their internal predictions of where the target is most likely to be. Because no visual cues were available, the only way people could do better than change is by learning any statistical rules underlying the target’s behaviour.

A number of different statistical rules (target movement behaviours) were examined in different conditions. In one condition, the target’s location was unpredictable, in other conditions it had favoured locations, in othes its location depended on where it was previously. People were not informed of what these rules were, they had to learn them directly from experience of where targets were in the past. Some of the questions that the paper addressed were:

  1. Do people learn that the behaviour of how the target moves over time? That is, do people learn about the statistical structure of their environment?
  2. Do people just make guesses about where a target will be (1st order statistics), or can they also make predictions about how it moves over time (2nd order statistics)?
  3. Can people use their observations of the target’s behaviour to make the best possible prediction of where the target will be next?

To find the answers, you can read the paper which is currently in press, and should be available to download soon.

Vincent, B. T. (in press) How do we use the past to predict the future in oculomotor search? Vision Research, doi:10.1016/j.visres.2012.08.001

Thanks to Alexander Tan (an undergraduate in the School of Psychology) who collected much of the data during a research apprenticeship.