The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9504

Title:Hector: the Use of Hyperschemas for Natural Language Generation in Exploratory Domains
Date: 1995
Abstract:The "user-friendliness" of hypertext systems is one of the many benefits that is often praised when discussing such systems. However, these systems are not wholly without criticism. In hypertext systems of significant size the problems of 1. repetitive or redundant information and hypertext links, 2. a lack of coherence between static pieces of information and hypertext links, 2. a lack of coherence between static pieces of information in the system and 3. the user getting "lost in hyperspace" trying to find the desired information are not uncommon features. Researchers in artificial intelligence, and more specifically natural language generation have sought to rectify these problems by developing systems that can tailor the textual content and followup links to be more sensitive to a user's needs. This thesis addresses these problems in relation to exploratory domains, i.e. domains where information is loosely structured, thereby allowing the user to be the drive behind the exploration of the information in the domain. From this class of domains the "features and topics of Edinburgh Castle" was the particular domain of knowledge chosen for exposition of the ;ideas in this dissertation. Through the use of hyperschemas, i.e. schemas that dynamically generate both explanation content and followup links to other explanations, implemented in the HECTOR (Hyperschemas for Edinburgh Castle TOuRs) system, it has been demonstrated that a coherent discourse of explanations and followups can be generated that is tailored to the interest of the user. The methods for systems development put forward in this thesis can therefore offer benefits, in terms of user-friendliness, that traditional hypertext systems lack.

[Search These Pages] [DAI Home Page] [Comment]