Mentor: Dialog Agent System for Mentoring and Conversational Role-Playing

Abstract

Abstract : The goal of this SBIR program was to provide authorable, dialog-enabled agents for tutoring and performance support systems. Users interact with agents who carry out strategies and goals and can engage in mixed-initiative dialog via a natural language understanding and generation system. Non-programmers can author new domains and scenarios and create new dialog agents. The dialog system is authorable by non-computational linguists. The system has two types of agents, Mentor agents and Conversational agents. The Mentor agent is a simulated subject matter expert (SME) that provides troubleshooting and problem solving advice. Mentor engages in a dialogue with trainees, helping them solve problems by taking them through logical courses of action and asking and answering domain-specific questions. Conversational agents are used for role-playing scenarios. The only real difference between the two agents is that Conversational agents do not have specific problem solving strategies. Both Mentors and Conversational agents have domain specific knowledge and access to a common sense knowledge base. This report describes the capabilities and limitations of results of this Phase II effort.