We conducted an experiment to test the hypothesis that self-explanation can be supported more effectively by means of natural language dialog than by means of menu selection of explanations. The study was carried out in a school in the Pittsburgh area and involved 78 students, 42 of whom satisfied all requirements. There was some evidence that students whose self-explanations were supported by dialog acquired better problem-solving skills than students who selected explanations from a menu. The evidence however was not sufficient to decisively confirm the hypothesis. Lessons learned include: it is more difficult for students to produce accurate free-form explanations than we expected and it may be better to delay strict requirements on explanation quality for students early in learning. It might be valuable to increase explanation strictness gradually. Also, knowledge-based natural language understanding capabilities push on hardware capabilities and have potentially negative consequences for software useability.
展开▼