The impact of bottom-up and top-down saliency cues on reference production


This study investigates to what extent visual saliency cues in realistic visual scenes cause speakers to include a redundant color attribute in their definite descriptions of objects, and in particular how such cues guide speakers in determining which objects in the scene are relevant distractors, and which not. First, regarding bottom-up cues, the results revealed that the presence of clutter positively affected the redundant use of color, but that the distance between a target and a distractor did not have an effect in this respect. Second, an effect of top-down saliency (i.e., whether a target’s type was mentioned in the instructions) was only partially borne out by the data. We argue that these findings are problematic for algorithms that aim to generate psychologically realistic object descriptions, since these generally select properties that help to distinguish a target from all distractors that are present in a scene.

Back to Table of Contents