Although the World Wide Web (WWW) and graphical tools for its use such as Mosaic are becoming increasingly popular for accessing and delivering information over networked computers, there have been relatively few explicit efforts to incorporate hypermedia authoring in the classroom. Science courses, including computer science and engineering (see, for example, [Aminmansour1993] or [Fowler and Fowler1993]), are being taught using multimedia tools. However, allowing students to design and build hypermedia documents as part of the normal curriculum is a relatively unexplored area and only some preliminary results have been reported [Weingrad et al.1993]. Some of this reluctance has been based on the steep learning curve associated with personal computer hypermedia languages. However, with the introduction of the HyperText Markup Language (HTML), sophisticated hypermedia documents can be created with only a small commitment to learning the underlying language. This results because most of the formatting is done via a browser's interpretation of simple but powerful HTML commands. This paper reports on a specific experiment in which large WWW hypertext projects were designed and written by students in a graduate course who had little-to-no exposure to Mosaic or HTML before the class began.
In previous years, Cognitive Architectures, the graduate computer science course at the University of Michigan in which the experiment was conducted, has included projects requiring an in-depth analysis of several representative cognitive architectures. This analysis was presented as a final paper at the conclusion of the term. However, this past year, the class (divided into five groups) developed independent analyses of all the architectures examined in the class using Mosaic and HTML to organize and present the information. A cognitive architecture may be loosely defined as a computational system that is designed to be applicable to a variety of general problems requiring intelligence. Although architectures may be characterized differently, the course focused on analyzing architectures according to their constituent components (e.g., a module for learning), the properties that arose from these components (such as the control strategy or knowledge representation scheme), general capabilities that the architectures possessed (problem solving, learning, moving about in the world, etc.), and the types of environments for which the architectures had been designed. One of specific purposes of using Mosaic was to develop a general framework for the analysis of cognitive architectures, not just for students but for researchers in Artificial Intelligence as well.
Outside of this very high-level description of categories for classification, the design of the documents was purposely underspecified by course professor John Laird with the intent of leaving many of the design decisions to the students, most of whom were being exposed to Mosaic for the first time. Thus, each group individually determined appropriate features for each architecture and how these features related to other architectures. These decisions were not always covered by the readings or lectures, forcing the students to consider possible alternatives and to determine the best connections based on their understanding of the material.
A major part of these decisions included determining the structure of the document itself. Work began in Mosaic after the first week of class and a thirty-minute introduction to HTML. As the semester progressed and new facets of architectures were introduced, the previous work had to be refined and updated. Thus, there was an almost continual reinforcement of the material from the very beginning of the class, unlike the traditional term paper which did not require assimilation of much of the course material until near its conclusion. In the process of building these documents, each group developed a unique strategy for organizing both the information and the underlying file structure. The completed projects differed considerably, ranging from 60 to 700 .html files with an average of about 5000 links per project.
This paper is organized as follows: the next section will address the anticipated advantages of using hypertext for this project, followed by a discussion of the different groups' approaches. The next section then presents a brief analysis of these approaches, based upon ongoing work by some of the students in the original class to combine the disparate projects into a single, comprehensive hypertext document. The final section focuses on the observed results of using hypertext authoring in the classroom.