Abstract
Temporal visual statistical learning (VSL) refers to the ability to extract temporal regularities from a succession of visual experiences. Little is known about how stimulus characteristics interact with VSL. In Experiment 1, we investigated how stimulus diversity impacts temporal VSL by presenting streams of face (female/male) and scene (indoor/outdoor) images presented one at a time to subjects. Unbeknownst to subjects, streams were composed of AB pairs of images, where A always predicted the appearance of B, which could be either same (e.g., male-male) or different subcategory (e.g., male-female), or different category (e.g., female-indoor). After familiarization, subjects completed a surprise forced-choice recognition task that pitted target against recombined foil pairs. When asked to detect "jiggle" events during familiarization, recognition was unaffected by categorical diversity. However, when asked to categorize events according to type of face or scene (female/male or indoor/outdoor), subjects were strongly affected by pair composition. Same-subcategory pairs were remembered much better than different-subcategory pairs, implying that categorization advantaged learning for more similar items. However, different-subcategory pairs required switching responses, creating response conflict that could have interfered with learning. In two additional experiments, we examined whether the same-subcategory advantage was due to response interference. Using only face images, in Experiment 2 we asked subjects to respond to gender or respond whether the face was the same or different gender as the preceding face. For same/different responders, we found the same effects of category even for pairs always associated with a response switch. In experiment 3, we flanked faces with congruent or incongruent category labels, reliably inducing response interference for some pairs but not others. We found no effects of response interference during training on subsequent memory for pairs. We conclude that temporal VSL is strongly influenced by categories of constituent stimuli, but only when those categories are task-relevant.
Meeting abstract presented at VSS 2018