David F. Dufty

Agent Technologies Designed to Facilitate Interactive Knowledge Construction

During the last decade, interdisciplinary researchers have developed technologies with animated pedagogical agents that interact with the student in language and other communication channels (such as facial expressions and gestures). These pedagogical agents model good learning strategies and coach the students in actively constructing knowledge during learning. This article describes computer technologies that have been developed during the last decade with tutors that attempt to facilitate deep comprehension (e.g., causal explanations, plans, logical justifications), reasoning in natural language, and inquiry (i.e., question asking, question answering, hypothesis testing). These tutors target high school and college students who learn about topics in science and technology. The primary example is AutoTutor, a system on the Internet that helps students compose answers to deep-reasoning questions and solutions to problems by holding a conversation. AutoTutors dialogue moves include feedback (positive, neutral, and negative), pumps for more information (“Tell me more.”), hints, prompts to fill in missing words, summaries, corrections of student misconceptions, and answers to student questions. Other learning technologies with agents include the Human Use Regulatory Affairs Advisor (HURAA); Source, Evidence, Explanation, and Knowledge (SEEK) Web Tutor; Interactive Strategy Trainer for Active Reading and Thinking (iSTART); Instruction with Deep-level Reasoning questions In Vicarious Environments (iDRIVE); and Acquiring Research Investigative and Evaluative Skills (ARIES). These systems have been tested on their effectiveness in facilitating knowledge construction. They also have uncovered insights on the prospects of designing agents to effectively communicate in language and discourse.

Automatic evaluation of aspects of document quality

Coh-Metrix is a web-based application currently in development that automatically evaluates text. It uses two central concepts from discourse processing: text-based cohesion and situation-model based coherence. Cohesion is the degree to which components of the text are linked. Coherence is the representation of the world that the text conveys. Our intention is for Coh-Metrix to eventually map the cohesion of a text to the background knowledge and reading skills of the reader. Coh-Metrix will then be able to give feedback to a writer about which aspects of the text are cohesive and which lack cohesion. This will enable the writer to determine which aspects of the text need to be improved. Applications of Coh-Metrix on document quality as well as other future directions for the development of Coh-Metrix are discussed.