Relational categories are notoriously difficult to learn. We studied the impact of comparison on relational concept learning with a novel word learning task in 3- and 4-year olds. We contrasted a no-comparison (single) condition and two comparison conditions. In the latter case, the set of learning pairs was composed of either close or far pairs (e.g., close pair: knife1- watermelon, knife2-orange; far pair: ax- evergreen tree, saw-log, for the “cutter for” relation). We also manipulated the transfer stimuli semantic distance (near or distant semantic domain, e.g., a scissor for a piece of paper in the close case, and a shaver for a face in the far domain case). The no-comparison condition led to random generalizations in the younger group only. Overall the close learning condition and the near transfer condition led to good performance. We discuss these results in terms of the role of semantic distance and how participants integrate stimuli depending on distance.

A common result is that comparison settings (i.e., several stimuli introduced simultaneously) favor conceptualization and generalization. In a comparison setting, we manipulated the semantic distance between the two training items (e.g., two bracelets versus a bracelet and a watch), and the semantic distance between the training items and the test items (e.g., a pendant versus a bow tie). We tested 5- and 8-year-old children’s generalization of novel names for objects. This study is the first one to study the temporal dynamics of comparison in a generalization task with eye-tracking data. The eye movement data revealed clear patterns of exploration in which participants first focused on the training items and compared them with each of the choice options. We also compared the search profiles for correct answers and errors. The results show that participants first found commonalities in the learning items, which they compared with each items in the solution set. This pattern is consistent with an alignment view of generalization.