Intelligent Dialog Tutor and Conversational Agents

This report describes an approach to develop intelligent and realistic computer-based training environments that are easily authored by non-programmers. It focuses on simulated agents who are capable of mixed-initiative dialog with the user through a natural language interface, either via speech or text. Trainees converse as they would in normal English discourse; asking questions, giving answers, and making comments or requests. The simulated agent generates appropriate responses to all of these inputs within any defined domain. This dialog capability will be realized in a simulated tutor and in simulated characters for mission training scenarios. The Intelligent Dialog Tutor provides individualized instruction based on what the trainee knows and does not know. After the lesson, the trainee can practice the newly acquired skills in realistic scenarios populated with Conversational Agents. The system is fully authorable, thus the Tutor can instruct and discuss any topic, including conceptual (e.g., leader skills) and procedural domains (e.g., maintenance tasks). Conversational Agents are easily created for new scenarios. Technologies include natural language processing, semantic analysis, discourse management, agent control architectures, a generic tutoring shell with a knowledge elicitation tool, continuous speech recognition, text-to-speech synthesis, and distributed object-oriented software for optimal Internet delivery and management.