Using tutoring patterns to generate more cohesive text in an intelligent tutoring system

If we want students to read, pay attention to and understand the text generated by an ITS, the text needs to (a) be organized into cohesive turns, (b) now in natural conversation patterns, and (c) have sufficient variety that neither syntactic structures nor pedagogical goals are repeated overly often. CIRCSM-Tutor v. 3 is a natural-language based TTS for cardiac physiology currently under development. Text planning for v. 3 will be accomplished by TIPS, a text planning engine based on current research in text generation. We describe how TIPS will achieve the goals mentioned above. We use transcripts of sessions conducted by expert human tutors to derive tutoring patterns for TIPS. Through the use of the TIPS planner, we expect CIRCSIM-Tutor v. 3 to generate longer, more complex and more natural text than previous work.