Abstract
Wu, Wang and Pomplun (2014) found that observers can use spatial dependency among objects for semantic guidance during natural scene inspection even when the scene gist is not available. This result suggests that observers are actively seeking semantic information to help visual performance. Here we investigate whether semantic association among objects is still used when there is no scene context, or even when using semantic information would not facilitate task performance. In Exp. 1, observers were asked to search displays of multiple object labels for the only repeated label. The objects referred by the labels could come from a single context set (ex. kitchen), from two context sets (ex. kitchen and bathroom) or from three context sets (ex. kitchen, bathroom and park). Subsequently, observers were asked whether a specified label was the repeated label or not. Interestingly, even though the task can be accomplished without accessing the meaning of each label, search accuracy improved and reaction time became faster when more context sets were shown. To test whether this bias was due to memory encoding which was indispensable in Exp. 1, we minimized the memory load in Exp. 2, which was a conventional search experiment with the target being indicated by either its label (label cue) or its image (image cue) before the search. The search displays were similar to Exp. 1 except that each label in the display only appeared once. The results show that, in the label cue condition, observers still had better search accuracy and shorter search duration (by up to 150 ms) when there were more context sets. Counterintuitively, this semantic association bias disappeared in the image cue condition. Our results demonstrate that the use of semantic association is not completely under voluntary control and operates differently in lexical and image processing.
Meeting abstract presented at VSS 2015