SGML-Based Markup as a Step toward Improving Knowledge Acquisition for Text Generation
Reva Freedman, Yujian Zhou, Jung Hee Kim, Michael Glass, Martha W. Evens · 1998
We are investigating computer-assisted methods for identifying plan operators at both the conversational strategy and surface generation levels. We are using standard-conforming SGML markup on our corpus in order to be able to process it mechanically. We are using C4.5 to identify rules of the form "when is goal x implemented with plan y?". We are currently testing these methods in the knowledge acquisition process for the text generation component of CIRCSIM-Tutor v. 3, a natural-language based intelligent tutoring system. Introduction CIRCSIM-Tutor is a conversational intelligent tutoring system (ITS) which uses natural language for both input and output. The text generation component of CIRCSIMTutor v. 3, which we are currently implementing, uses a two-phase architecture consonant with the consensus architecture described by Reiter (1994). A global, top-down tutorial planner chooses and instantiates a logic form for the system to say based on available information The tutorial plann...