New scene representations are learned more effectively during active than passive tasks
Victoria Nicholls1, Maxim Spur2, Erwan David3, Melissa Vo1

1 LMU (DE)
2 ENIB (FR)
3 TU Le Mans (FR)

When we search for objects in a room, we use our previous knowledge of scenes to direct our attention to locations where the object could be. This scene knowledge, a.k.a. scene grammar, follows a hierarchical structure (Võ, 2021). Currently, it is not clear how we learn scene grammar. For instance, do we simply learn through repeated exposure to objects being in the same location or is a more active task, such as search, required to successfully learn object identities and locations? In this study, we aimed to determine how adults learn a new, artificial scene grammar. Participants encoded objects through either a passive viewing or an active search task. The passive viewing task highlighted objects in the scene which participants gazed at one after another. In the active search task, participants were given a target object which they searched for. To ensure that participants had no previous scene grammar for the objects or the scenes we presented them with artificial objects in artificial virtual reality scenes.  Participants then performed two recall tasks. First, an odd-one-out task where participants were presented with three objects and asked to choose which one was the odd one out. Secondly, participants were given a scene reconstruction task where participants had to place the artificial objects in the correct locations within the scenes. Eye tracking and behavioural responses indicated that participants learned the object locations faster and, in more detail, when they encoded objects via the active search task compared to the passive viewing task. However, participants only showed superior memory regarding which objects were grouped together in the rebuild task and not the odd-one-out task. These results suggest that participants learn new scene grammar more effectively through an active search task and only recall the new scene grammar representations when required for the task.