Explanation Content and Construction during System Learning and Troubleshooting

Explanations have become increasingly recognized as central to learning and intelligence. To explore the content and construction of human explanations, detailed protocol analyses were performed on the explanations that subjects produced while learning about a fictional system. During the system learning phase of the experiment, subjects received a series of diagrams for a system and then explained how it worked. During the troubleshooting phase, in the absence of the diagrams, subjects explained relations between problematic symptoms and their faults. To examine the evolution of explanations with learning, subjects explained the same system diagrams and the same symptom-fault relations multiple times. Explanations varied considerably between system learning and troubleshooting, indicating that different goals produced different explanations. During system learning, subjects focused almost exclusively on functional explanations, exhibiting no trend toward deeper mechanistic explanations over time. To construct these functional understandings, subjects primarily instantiated existing knowledge in memory, specializing it to explain the purposes of individual components. During troubleshooting, subjects still focused on function but included a much wider variety of functional information in their explanations as well as higher proportions of topographical and mechanistic information. Rather than focusing on individual components, subjects focused on qualitative relations between components as they