The role of conceptual structure and background knowledge in category learning
Johnson, Matthew Carl
MetadataShow full item record
Two experiments were conducted in order to determine whether background information acquired by reading from text differentially influences category learning relative to when no background information is provided. Experiment 1 was a control study that compared short and long versions of text containing information that describe the characteristics of different plant features (e.g., roots, stems, leaves, and flowers) and how each is able to adapt to the characteristics of desert and mountain environments. Seventy-two participants sorted eight drawings of plants into two categories (desert and mountain plants) and read either a short list or a longer, more elaborate text describing the characteristics of plant features. Then after reading, they answered comprehension questions over the text until they mastered the information, and then they re-sorted the plants again. The results indicated that learners applied what they had read when resorting as evidenced by fewer errors relative to initial sort patterns. Experiment 2 compared the learning of linearly separable and non-linearly separable concepts for groups of participants that either received no background information (no text), read background information from a text to a high level of mastery (comprehensive text), or read at their own discretion prior to learning (available text). After participants completed one error free run through the eight training stimuli, they classified old training items and eight new transfer items. The results indicate that requiring participants to fully comprehend the text (the comprehensive text group) facilitated learning of the linearly separable concepts, but not non-linearly separable concepts. This finding is consistent with the assumption that learning is enhanced when the items to be learned do not violate background knowledge, as was the case for the comprehensive text group who learned the linearly separable concepts. In addition, transfer performance for the available and comprehensive text groups was driven more on what they had read relative to how similar the items were to past examples. However, exemplar similarity was predominantly used by all groups after learning non-linearly separable concepts. These findings were interpreted as supporting a mixed representational model that accounts for both exemplar similarity and background knowledge.