A comparison of tasks for constructing the category space of natural scenes

Similarity is a core construct underlying various prominent category theories. This study compares the outcomes of two classic similarity judgment tasks—a same-different judgment task (implicit similarity task) and a spatial arrangement task (explicit similarity task)—for natural scene categories. We ask whether or not the two methods produce the same category space, that is, the similarity structure within and between categories. The same natural scene set (four categories: beach, city, highway, mountain; from Torralbo et al., 2013) was used for the two tasks. The same different judgement task (N = 218) required subjects to respond whether two side-by-side presented natural scenes belong to the same category or not. The similarity of the scene pair is implied by the response times of the judgement. The spatial arrangement task (N = 37) asked subjects to sort 24 scenes such that more similar scenes are closer scenes to each other. Two different sorting lists were included: within category (i.e., 24 images were all from the same category) and between category (i.e., six images per category). The multidimensional scaling (MDS) results were computed for both tasks, and the image distances were derived based on the MDS coordinates. We correlated the arrangement of each trial (i.e. 24 scenes) with the average MDS from the same-different task. Results showed that two tasks were significantly albeit weakly correlated for both within and between category distances (within category: r = .10; between category: r = .18). This suggests that while these implicit and explicit tasks might construct approximately similar spaces, there are still considerable differences, raising the question of whether we can use the tasks interchangeably to learn about the structure of natural scene category space.