There have been decades of efforts on research and development of intelligent tutoring systems (ITS). Many tutoring systems provide rich media content and allow students to interact with content in many different ways, such as answering multiple-choice answer selection, dragging and dropping objects, rearranging objects, assembling objects, and so on. Intelligent systems assess students’ performance from the data collected from the interactions and then adaptively select knowledge objects and pedagogical strategies during the tutoring process to maximize the learning effect and minimize learning cost. Delivering content with conversation is always attractive to content authors and students. For example, when a piece of knowledge is delivered through a text message, would it not be more interesting to have a conversation between a “tutor”, human or machine, and the student to talk about what is in the text? Research has shown that delivering content through conversation is much more effective than a text. Unfortunately, creating conversational content is difficult. First, in order to have a natural language conversation with a student, the machine has to be able to “understand” the student’s natural language input. This involves a research field called “natural language understanding.” There isn’t a perfect natural language algorithm that can really understand user’s free-form speech. Secondly, preparing tutoring speeches for conversations is hard. The essential difficulty is that authors will need to consider the appropriate amount of responses to an infinite possibility of student input. Additionally, it is hard to create and test conversation rules. Conversation rules decide the condition under which a prepared speech is spoken. Since the tutoring conversations often go with other displayed content (e.g., text, image, video) conversation rules need to take into account all activity within the learning environment, in addition to the natural language inputs from students. The rule system varies because different environments generate different activity. Creating and testing the rules is also time-consuming. Other difficulties involve talking head techniques (e.g., speech synthesizing, lip synchronization, emotion, gesture), speech recognition, emotion detection, and so on.
The AutoTutor team at the Institute for Intelligent systems (IIS) at the University of Memphis has been working in this direction since the 1990s and has been providing solutions to overcome the difficulties in conversational ITSs. About a dozen conversational ITSs have been successfully developed in the IIS, including a computer literacy tutor, conceptual physics tutor, critical thinking tutor (OperationARIES!), adult literacy tutor (CSAL), electronics tutor (ElectronixTutor), etc. A team at the National Taichung University of Education has developed a Chinese language tutor.
AutoTutor helps students learn by holding deep reasoning conversations. An AutoTutor conversation often starts with the main question about a certain topic. The goal of the conversation is to help students construct an acceptable answer to the main question. Instead of telling the students the answers, AutoTutor asks a sequence of questions (hints, prompts) that target specific concepts involved in the ideal answer to the main question. The AutoTutor systems respond to students' natural language input, as well as other interactions, such as making a choice and arranging some objects in the learning environment. This tutorial focuses on the authoring process of AutoTutor lessons, including discourse strategies in AutoTutor dialogues and trialogues, conversation elements, media elements, conversation rules, and template-based authoring.
