Finding abstract commonalties of category members

To form a coherent conceptual category and use it for inductive inference, the cognitive system needs to discover commonalties among different objects. How does the system accomplish this task? This study compares two of the most commonly used functions of categories–classification and feature inference–and examines their effectiveness in finding abstract commonalties of category members. The results from two experiments show that a classification task is not very useful for abstraction. In contrast, a feature inference task is advantageous in extracting abstract commonalties. However, this advantage is limited. Finding abstract commonalities becomes burdensome when category labels are absent in the feature inference task. These results underscore the importance of category membership information for abstraction. It is suggested that this advantage comes from the fact that category labels help form structured representation and facilitate structural alignment.