Hierarchical networks 145
History 146
Gagné 148
Task Analysis as Problem Space construction. 150
Means and Gott 150
Kurland 150
Hawk Mach-III 150
HAWK MACH-III and Qualitative Simulation 151
The Simulation 152
Hypergraphs 153
Hypertext 154
Evaluation 155
Design Environments 155
IDE and Notecards 155
Janus 157
Procedural Hierarchy of Issues (PHI) 158
Instructional and Cognitive Problems 161
Instructional Software Design and Production 162
Guidelines for Instructional Design
o Analyze the design space and decompose the knowledge and skills into bite-sized chunks of a knowledge network.
o Describe the structures and functions in terms of text, pictures, icons, and simulations.
o Analyze the knowledge and skills of your audience into a knowledge network.
o Determine your purpose and goals.
o Pick a representative case or model.
o List your arguments pro and con; and the evidence for them.
o Analyze your own role as a mediator of learning.
Putting these guidelines into practice is anything but easy. The challenge of teaching lies precisely in the ambiguity about doing all of these things systematically, and then using the final product of the design in a flexible, dynamic and interesting way. To combine both the rigor of systematic design and the opportunism of making best use of all available resources in the design calls on all the professional skills and ability of the best teachers. It is unlikely that any computer-based system, using artificial intelligence, hypertext, and any other accessible technology, will replace the intelligent flexibility of a human teacher in the foreseeable future. However, the powerful assistance that these technologies can provide simply cannot be ignored. They offer innovative situations for teacher-student cooperation and interaction. The real challenge for instructional design in the era of advanced technologies is devising ways to engage teachers and learners in environments where knowledge is already externalized and predigested, waiting to be used.
Hierarchical networks
One of the most powerful tools to have come out of the computer science field, particularly from teh AI area, is the understanding of how to segment knowledge into hierarchical networks and even simpler trees. This technology is so widespread that it is taken for granted by great numbers of computer hackers. Its most prominent exposition came in the work of Newell and Simon (1972) who proposed the notion of a problem space, an analysed collection of the goals, plans, and facts pertaining to a problem. The strategy of decomposing knowledge was carried forward to finer degree in much work to follow on semantic nets (Collins and Quillian, 1969) and in the early intelligent tutroing systems (Carbonell, 1971, Grignetti, 1974). The work on BUGGY provided a new generation of analysis based on bugs and GAO (Generalized And - Or) graphs ( Brown and Van Lehn, 1980). At the same time Soloway (1978) produced his bug libraries and goal-plan hierarchies. The power of these generalized graph structures for instruction is only beginning to be explored. Although analyzed hierarchies exist for some small domains ( especially programming languages like LISP, PROLOG, and PASCAL) similar analyses have been carried out for only a small number of other domains. If such an analysis were widespread, it is not only thinkable that the whole curriculum from elementary grades to college could be revamped and streamlined, but it is entirely possible that individual learners' job of learning all this stuff could be made much easier and better integrated. The vision of this global interconnected network of knowledge raises the distinct possibility that each child and adult could learn much more than she or he can learn under today's system. Of course, the notion of this large connected network of ideas should immediately raise in your minds a vision of hypertext assisted learning (HAL) and navigation.
Narrative Structures
Rand Spiro
Denis Newman's study with children
History
One of the first hypertext systems commercially available was really an instructional design environment, influenced by the work of M. D. Merrill. It was called TICCIT, and it put into actual practice on a (then) large computer system the principles that Merrill developed within a framework he called Component Display Theory (CDT). Although the general issues of nodes and links and hypertext structures were not explicitly or formally discussed during the period of TICCIT's development (although see Stone, 1972? for some early comments) the general issues of navigation and control were confronted quite directly with some remarkable innovative solutions that appear quite familiar in hindsight. An examination of the theory and implementation will lay the groundwork for more modern perspectives.
CDT comprises three major parts (Merrill, 1988): a classification system of student performance (remembering, using and finding), subject content (concepts, skills and principles), presentation techniques ( rules, examples, and practice), and links among the three. It is based on a general assumption that different kinds of objectives are best learned with specific kinds of presentation techniques and materials. In practice, the theory leasds to the presentation of general rules followed by more specific examples, followed by extensive practice on even more examples, with feedback and help or advice adjusted to the student and under the student's control.
However, much of the theory consists of overly general rules that have to be adapted (sometimes radically) to any particular environment. For instance, a rule about fading generally argues that the amount of help should be faded out as learning progresses. While anyone might agree with that, putting it into practice raises some real problems about what has been helped. For instance, in teaching division using a computer-based environment, the help that the computer and teacher give may well be on the lower level multiplication facts rather than on the division algorithm itself. As this algorithm is learned, and division moves on to multiple digit and decimal numbers; the system could actually increase its support of the lower level components of the division algorithm (subtracting, multiplying) while letting the student focus specifically on how to deal with the remainder and "bringing - down". So, in a very real sense, help is continually faded in as well as out as the lesson is learned. Since, at the end of the "lesson", it is now assumed that a student knows a particular component of a lesson, she may be helped indiscriminately on that component while she is helped sparingly on the new and more difficult materials. Similar "picky" arguments can be made about many of the components of CDT theory.
Instead of dealing with these rules and the "theory" of CDT, it may be more useful to examine how TICCIT implemented real insights into sharing the computational prowess of a computer with students in a learning environment.
To begin with, TICCIT encourages a great deal of learner control. At the learner's disposal are a number of powerful keys to navigate through the lessons in an orderly and predictable way. Overall, the lessons are always structured within a hierarchy that begins with a general rule, descends to particular examples, and then moves to specific practice. In fact, TICCIT makes explicit use of this hierarchy in a primitive student model that encourages students who are having difficulty to move back up this hierarchy. This model is invoked by pressing an Advice key. The hierarchy is explicitly available on the keyboard with three keys marked "Rule, Example, and Practice". The hard and easy keys modify all three parts of the hierarchy by having alternate forms of each available. These keys actually bring up individual frames of text or graphics that provide separate alternatives to the current frame. A help key points out salient facts or procedures.
In addition to all of these splendid support components that reduce the likelihood of a student's losing her way through the maze of instruction, TICCIT pioneered the use of a map or hypergraph. A graphic overview of each lesson could be brought up on the screen simply by pressing the Map key. From this Map, the student can choose which lesson to study and which rules, examples, and samples to work on. Generally these maps provide a prerequisite hierarchy for the current lesson, and show the student's state of progress through the hierarchy. Creating this very rigid and detailed structure is an enormous amount of work, and clearly not all lessons authored in TICCIT made use of these features. However, the availability of this kind of detail is theoretically very useful. Modern hypertext systems would implemnt this hierarchy by using a prerequisite link and careful analysis to work out the proper sequence of prerequisites. Our understanding ofhow to do this properly underlies everything about the structure of a curriculum.
If you think of teh grand overarching structure of a curriculum, particularly through the first eight grades of school, ti is utterly amazing that such a complex structure could work. How is it that we are able to analyze the structure of all the many things we know in such a way that we can begin with such elementary skills (for instance) as the number facts and continue with addition and subtraction all the way to calculus without missing a step? Is this sequence really as tidy as we think it is? Could there be a better sequence? Should everyone go through the same sequnce, or are there other sequnces of instruction that would be better for individual students?
These are really the basic questions underlying teaching and learning; and it is instructive to look at the ideas behind a prerequisite hierarchy for answers.
The use of this prerequisite hierarchy links TICCIT to the pioneering work of Robert Gagné, to which we turn next. However, before we leave Merrill's CDT, it is necessary to acknowledge the widespread influence the many applications created by Merrill have had. While CDT itself is opaque in multisyllabic headings and terminology, many applications produced under its guise have pioneered simple and effective techniques for instruction. More recent examples include more emphasis on interactive demonstrations, microworlds, and dynamic simulations that are difficult to fit into the strictures of CDT. Of course, one may always interpret them as very large examples or exercises, overshadowing the rule component, but this really seems to abuse the elegance of CDT theory. Instead, it may be better to retreat from the complexity of CDT to something more straightforward. Other examples of CDT - like applications have centered on poetry and literacy skills, with much more emphasis on rules and long descriptive passages of text. Again the theoretical strucures of CDT are strained by the specific needs of textual representations. These two knowledge representation forms, text and graphic simulations, appear to be radically different, each with its own needs, yet both lead naturally into learning hierarchies.
The fundamental differences between textual and graphic organizations of subject matter makes it seem natural to focus on text and simulations; on presentation and practice; on descriptions and problems; on knowledge and skills; and on declarative versus procedural forms of knowledge as complementary systems to guide the creation of instructional design. The principle that applies to both forms of knowledge is decomposition: both declarative and procedural knowledge can be broken down into simpler primitive concepts; and these concepts can be arranged meaningfully into networks and simple hierarchies.
Gagné
The concept of a "learning hierarchy" has been proposed and analysed in considerable detail for a great many intellectual skills. Essentially, Gagné (1979) proposed that all intellectual tasks and skills could be decomposed into a learning hierarchy of successively simpler skills. Figure 6-1 is an example of a learning hierarchy that Gagné proposed for subtraction. At the top of the figure are teh more complex skills that may each be the objective of a single lesson. Below that are the prerequisite skills that "enable" or make it possible for a student to perform the higher skill. The implication of this hierarchy is that the lower skills should be taught first, as prerequisites for the higher skills. The learning hierarchy thus provides a basis to guide the sequence of a curriculum, or at least to guide the planning of examples and exercises,
Gagné proposes that not just intellectual skills, but verbal information, attitudes, and motor skills have this hierarchical structure of prerequisite skills. For instance, he argues (p. 114) that swimming involves the coordination of many separate part-skills: breathing, kicking, stroking, floating, each of which can be practiced independently. Curiously, he also argues that an overview of all the skills put together may be necessary, even at the beginning to set up the proper learning objectives.
Task Analysis as Problem Space construction.
Means and Gott
Kurland
Hawk Mach-III
Advanced technologies, including artificial intelligence (AI), hypertext, and natural language processing (NLP), are transforming the Mind/Machine Interface. Their use is guided by the three natural means of communication between people: saying, coaching, and showing; as metaphors for using advanced technology interfaces. The most complete example of technologies that implemetn these insights through Hypertext and qualtiative simulations is the Maintenance Aid Computer for HAWK--Intelligent Institutional Instructor (MACH-III). This is the largest and most successful implementation of an ITS to date (Psotka, Massey, and Mutter, 1988) . MACH-III was developed by Bolt, Beranek, and Newman (BBN), to provide training in organizational maintenance of the main radar of the HAWK air defense guided missile system . Its core is a huge qualitative simulation of the radar. The complexity of the simulation and the troubleshooting problem space demand a unique hypertext interface, whose structure and function are only beginning to be understood. Some preliminary evaluation results from the U.S. Army Air Defense Artillery School (USAADASCH), Ft. Bliss, Texas are beginning to show its effectiveness.
Our goal is to develop computerized training systems that can provide clear improvements over traditional training. Improved training relies on a steady advance in mind/machine technologies. Unless the training machines can communicate with the trainees, little training can be expected. Three kinds of communication that mainly occur between teacher and trainee are through speech, feedback or critiquing, and interactive demonstrations. If a computer is to undertake these communication functions it obviously needs natural language for speech, and qualitative simulations for demonstrations. However, critiquing can be effectively performed by either communication medium, or by hypertext. It may in fact be uniquely attuned to hypertext systems. These three technologies (qualitative simulations, hypertext and natural language processing) epitomize the three most promising areas for potential advances in mind/machine technologies. Of the three, qualitative simulations are furthest advanced. Hypertext systems are finding wide application and progressing steadily (Shneiderman and Kearsley, 1989). Natural language processing technology is only recently beginning to be exploited, particularly with the advent of fast Prolog systems. It may well be the case that once NLP is well-established it will completely take over from current hypertext systems, so that hypertext will be only a transitional phenomenon. However, hypertext may include powerful browsing systems that have graphic interfaces (hypergraphs) of conceptual structures. These hypergraphs may have robust critiquing properties ( Collins, Brown, and Newman, 1988) that will work to complement the strengths of simulations and natural language.
HAWK MACH-III and Qualitative Simulation
The complexity of a radar is stunning. In the Army, trainees (usually highschool graduates) have to learn to maintain it from thick folio documents affectionately known as "branch and get lost" manuals. These are basically engineering documents that divide the radar along design lines, largely inappropriate for functional troubleshooting. The challenge of MACH-III was to create an entree into this complexity and make it manageable. Issues connected with the training design are detailed in Psotka, Massey, and Mutter (1988). Interface issues dealt with complexity in three general ways. The qualitative simulation provided the heart of the approach. But even its complexity was staggering. In order to hide some of the complexity and reveal it slowly (Dede, 1989) the fault isolation procedures were separated out and placed in a hypergraphic browser that provided an overview of the problem space. Finally, several navigational aids, using text and hypertext explanations were created so that trainees could find their path and not get "lost in hyperspace". Managing the complexity of this knowledge with these three levels of symbolic abstraction ( simulation, hypergraph, and hypertext) allowed novices to select their most congenial entrée into the knowledge base. Each navigational aid appears to represent a different level of symbolic abstraction, a different emphasis on declarative versus procedural representations, and a different (sometimes overlapping and sometimes complementary) functionality in the ITS.
The Simulation
Qualitative simulations make inspections of the causal relationships among components much more direct than any physical device can allow. Following the general design guidelines introduced by the graduated models of QUEST (White and Frederiksen, 1986), the direct views of simulated radar screens make it possible to manipulate complexity of representation and provide extensive, repetitive drill and practice on the most difficult procedures to learn. By having integrated and graduated perspectives on a domain, the more familiar structures and relationships can be used as a scaffolding to support and hang on more complex, abstract and hidden structures and functions. Careful design of the MACH-III simulator was intended to provide a symbolic model that could be used to extract inferences about the symptom-cause troubleshooting complex of knowledge. In effect, the simulation captured much of the explanation that should have been in the fault isolation manuals, but is not there. It also provided an environment where troubleshooting steps could be demonstrated, and troubleshooting exercises observed. In effect the simulation provides the procedural environment for trying out skills and honing them to perfection.
Because of the complexity of this ITS and the heavy computational burden it placed on the Symbolics © computers, it was particularly important to create knowledge representations that are compact and run efficiently. The object oriented graphics system was made efficient with an abstraction hierarchy that dealt not only with the component electronic cards and subsystems, but explicitly with loop and chain structures, and with information, noise, and signal flow. These relationships were visualized with dynamic pipelike wires with visible flow patterns. Every component had popup menus that provided context sensitive links to explanations and routine actions that were needed during troubleshooting.
The great deal of original creative labor that went into designing this system underlines the scarcity of general design principles. Too little work has been aimed at developing new representations for information and relationships. There are several important categories: representing justifications, consequences, and causal mechanisms. Of most obvious importance is the development of structures that encode and make explicit causal connections, and decompose systems into simpler causal structures where functional relationships are more apparent.
Qualitative simulations provide convenient vehicles for creating systems with both meaningful structural and functional components. Instead of a textual description of terminology and its interrelationships, the structure is described visually. Functional relations can also be described by using animation, color, arrows, and textual descriptions. However, visual descriptions of both structure and function are notoriously concrete: it is difficult to obtain the right level of abstraction without the use of conceptual descriptions in a textual and hierarchical form. Furthermore, it is difficult to create visual descriptions at different levels of abstraction in a truly hierarchical format. Functional and structural relations must be compressed or eliminated in arbitrary ways that defy accurate concrete representations. So, it is only when conceptual (textual) and concrete (visual) representations complement each other that adequately faithful models can be created.
Hypergraphs
An interactive troubleshooting tree browser (hypergraph) provides a somewhat parallel interface to the underlying mental model of the radar. This hypergraph not only promotes browsing (skimming through for the nodes of interest) but it too provides a graduated entrée into the simulation. The hypergraph is at a level of graphic abstraction higher than the simulation. It also contains much different information. The hypergraph organizes the procedural knowledge of the troubleshooting steps into an abductive hierarchy that begins with the most general and indicative symptoms, leads down to possible malfunctions, then branches to actual troubleshooting tasks, and finally points to the devices that need to be tested. The hypergraph is thus an explanatory complex that relies intimately on the causal flow embodied into the simulation to select the best possible path through the hypergraph. The hypergraph is not a decision tree; it is an explanatory, abductive hierarchy. Although it is a separate entity from the simulation, it is constructed in a thoroughly principled way so that the interpreter connecting the two (the hypergraph and the simulation) is highly general and needs no structural knowledge to make the connection. The hypergraph is an enormously powerful navigating device.
There are several levels of navigation assistance embedded in the hypergraph. In the most direct assistance, nodes in the hypergraph are grayed out after each visit. Each node has also attached a set of popup context-specific menus to perform a range of functions: expanding the node, engaging the simulation, finding components, or providing hypertext explanations. More complex assistance is provided by an active Advisory/Critiquing function. By interrogating the state of the simulation, the Advisor can activate nodes in the hypergraph that point to the most likely symptoms, malfunctions, troubleshooting steps, or troublesome devices. The Advisor can also monitor the hypergraph for actions and implement them on the simulation, thus using the simulation to help trainees navigate through the hypergraph. Finally, the Advisor can move to a higher level of abstraction by converting the connectivity of the simulation into explanatory hypertext.
Hypertext
Hypertext provides a text based system that goes beyond text to include graphics, video, and sound (hypermedia) as well as links, crossreferences, and network or lattice structures (hypergraphs). Many expert systems confound the structure of knowledge in their systems by combining many different knowledge components into any one rule. This makes it not only difficult to update and modify, but also it is difficult to get direct access to the information for explanatory or control purposes, (Clancey, 1986). These are really flat semantic structures. Perhaps, when the rules are coherently ordered and systematically constructed for a particular problem space, there may be an implicit hierarchical semantic structure. Usually, however, this implicit structure is not fleshed out and certainly not used for explanation purposes. There are very few attempts to map out the hierarchical structure of a problem domain, although one of the earliest attempts at constructing ITS just for this purpose. The hierarchical organization of the simulation and the hypergraph lend themselves naturally to an explicitly hierarchical explanation system in MACH-III that is unique among existing ITS.
Evaluation
How effective is this complex interface for providing novices an entrée into the even more complex world of radar troubleshooting? We at least have some preliminary evidence that it is effective (Kurland, Granville, and MacLaughlin; 1989; Kurland, Psotka, and Acchione-Noel, In press). A group of 11 Army soldiers and Marines used MACH-III in their regular course on the receiver and transmitter of the HAWK radar. Compared to another group without the ITS their performance was one standard deviation higher (t = 1.88; df = 20; p = .04). Naturally, these are only preliminary, informal findings, and the official Army training evaluation agency will probably issue a formal report by the end of 1989. Preliminary results from the ongoing evaluation are in keeping with these exploratory results, again finding superior performance on a number of measures by groups trained with this hypertext interface to a qualitative simulation.
Design Environments
The two prototypical design environments that are also hypermedia systems to be discussed here are IDE: the Instructional Design Environment built on top of Notecards; and Janus, built on the Document Examiner.
IDE and Notecards
The Notecards (Halasz, Moran, and Trigg, 1985) system is a hypertext environment designed to aid in the collection, structuring, and analysis of textual and graphical information. The metaphor upon which Notecards is based is that of notecard (e.g., 3 by 5 cards) and fileboxes. A Notecard is intended to contain a small, single, idea-sized chunk of information, in either textual or graphic form. A collection of Notecards can be arbitrarily linked together to form networks that convey the relationships among the ideas stored in the various Notecards.
Each link in the network has a type associated with it indicating the kind of relationship that exists between pairs of ideas. Multiple links in many directions are supported. The links are visible within each notecard as a hotspot title or icon that is mouse-sensitive and pops up into the full text or graphic when it is buttoned. There are many card types, each
supporting one of the full range of functionalities available on the lisp machine.
NoteCards offers designers a tool with which to organize, manipulate, and structure their ideas. Furthermore, it permits the designers to analyze the structure of their ideas, revealing not just the content of their knowledge, but also the way in which facts interrelate. It gives the designer the ability to view the contents of a knowledge domain from any one of a number of perspectives. The multiple card types of NoteCards - editor, text, graph, sketch, simulation, student model, instruction, etc. - organize and provide a common user interface to many functions. We envision that a NoteCards design in a frame-like environment will lead to automatic construction of objects and simulations, and we are actively exploring this view in a series of design and development environments arising from Notecards. We have called this environment IDE (Instructional Design Environment) and it is available on a research basis (Russell, Moran and Trigg; 1988). We ave developed some examples and prototypes to explore how to use hypertext in multilingual courses for the 97E military occupational specialty (MOS) in English, French, German, and Spanish to be taught interactively using this environment. An example of this structure for German is given in Figure 2.
FIGURE 2: A sample IDE screen for German, showing hypertext hotspots, graphics, and
animations.
Janus
Many hypertext systems can be seen as fundamentally programming and design environments. Authoring and designing instructional systems are also fundamental teaching and training activities, so it is most useful to begin this section with an overview of some exemplary work by Gerhard Fischer, Raymond McCall, Anders Morch of the University of Colorado, Boulder in a project called JANUS: Integrating Hypertext with a Knowledge-based Design Environment. They have worked for a number of years on the development of computer systems, including hypertext systems, to support designers in a wide range of fields, including architecture and software design. Their overarching approach is to create computer systems that can support and even cooperate with human experts who are learning their skills on the job, learning by doing. In their systems, learning by doing is tightly coupled with learning on demand. This close connection ensures that new knowledge is made available when it is most likely to be learned, when the user needs it to do her job.
Their guiding concern has been, how can hypertext help designers in finding information useful for activities they are currently engaged in? Designers are not interested in exploring hypertext information spaces per se but rather in obtaining information to solve problems or accomplish tasks they are dealing with.
Even expert designers can no longer master all the relevant knowledge, especially in technologically oriented design, where growth and change of the knowledge base are incessant.
Conventional retrieval also has problems in effectiveness and efficiency. Its effectiveness is limited by a seemingly unavoidable tradeoff between the two standard measures of effectiveness: precision and recall. The former denotes the fraction of retrieved items which are relevant; the latter, the fraction of relevant items in the information base which are retrieved. No conventional approach can come close to finding all those items and only those which are relevant. Typically, users must wade through large amounts of retrieved items to find small amounts which are useful.
JANUS substantially reduces these problems for design by integrating a knowledge-based system with an issue-based hypertext system. Our focus will be on the issue-based hypertext.
Procedural Hierarchy of Issues (PHI)
The Procedural Hierarchy of Issues (PHI) approach [McCall, 1987] creates the framework for the issue-based hypertext by analyzing the process of reasoning, problem solving, and argumentation into a special structure relating issues, answers and arguments.
A second crucial process is the recursive decomposition of issues into subissues. This characteristically results in a tree-like structure of an issue with subissues,
ISSUE 101:
What should the arrangement of the various components of the kitchen be?
SUBISSUES:
102: What are the various components of the kitchen that need to be arranged?
103: What should the locations of these various components be?
SUBISSUES:
104: What should the locations of the various architectural features be?
SUBISSUES:
105: What are the various architectural features?
106: What should the location of the walls be?
107: What should the location of the doors be?
108: What should the location of the windows be?
109: What should the location of the plumbing be?
110: What should the location of the equipment area be?
SUBISSUES:
111: What is the longest usable wall length?
112: What should the location of the eating area be?
113: What should the circulation pattern be?
114: What should the locations of the various components of the equipment area be?
SUBISSUES:
115: What are the components of the equipment area that need to be arranged?
116: What should the location of the cleanup center be?
117: What should the location of the cooking center be?
130: What should the location of the storage center be?
118: What should the location of the preparation center be?
119: What should the locations of the various components of the cleanup center be?
123: What should the locations of the various components of the cooking center be?
127: What should the locations of the various components of.the storage center be?
131:What should the locations of the various components of the preparation center be?
FIGURE 3. A Hierarchy of issues with subissues.
A real strength of PHI is that it completely and accurately models the task structure of the design process and the information useful for tasks.
The link labels in screen displays also tell designers
1) of the existence of information useful for current design tasks,
2) the basic nature of this information and
3) how to retrieve it (by clicking on it).
This allows retrieval of complex substructures of the hypertext network using numeric identifiers, indexed terms or substrings in combination with descriptions of network structure. VIEWPOINTS is implemented in HyperCard on the Macintosh.
Human Problem-Domain Communication
To shape the computer into a truly usable and useful tool, users must be able to work directly on their problems and tasks. The goal of human problem-domain communication (Fischer et al., 1988) is to eliminate computer-specific programming languages and instead to build layers of abstractions with which domain specialists--such as kitchen designers--can feel comfortable.
Construction kits are domain-specific programming languages that help users to formulate solutions to complex problems and that allow them to create complex artifacts without having to master the many details inherent in general purpose programming languages.
Design environments, such as CRACK, combine construction kits with critics. Critics use knowledge of design principles to detect and critique partial solutions constructed by the designer. Critics in CRACK are state-driven condition-action rules which are triggered when non-satisficing partial solutions are detected
Evaluations of these systems together with studies of designers by ourselves and others show that integrated support for construction and argumentation is essential for full support of design. Schoen's theory of reflection-in-action states that constant alternation between action and reflection, i.e., between construction and argumentation, is the hallmark of good design. The shift to argumentation occurs when designers need to evaluate what has been constructed, are stuck in problematic situations or need to think about what to do next.
Designed artifacts "talk back" to designers as they are being designed. This dynamic interaction is essential for good design. The experts rely on it,and novices are notoriously insensitive to it. Good environments promote and support efficient talkback by providing the tools for reflection and analysis: hypergraph browsers, ;audit trails, critics, and deep explanations tied easily and conveniently to the design environment.
JANUS' concept for integrating CRACK and VIEWPOINTS derived from the observation that CRACK's criticism is really a limited type of argumentation and can be mapped directly into PHI form. In particular, the construction actions can be seen as attempts to resolve issues.
The fact that CRACK's criticism can be put in PHI form shows that its knowledge-based critiquing mechanism actually connects construction with argumentation--albeit very limited argumentation. This means that CRACK and VIEWPOINTS could be coupled by using CRACK's critics to provide the designer with immediate entry into the exact place in the hypertext network where the argumentation relevant to the current construction task lies.
When users enter JANUS-VIEWPOINTS from JANUS-CRACK they are brought into a section of the PHI issue base determined by and relevant to their current construction situation. They do not have to hunt for relevant information. Their point of entry into the hypertext network will contain relevant information. But since the argumentation on an issue can be large and complex, they can use the initial display of relevant information as the starting place for a navigational journey through the issue base. Each traversal of a PHI link will take them to additional relevant information which argues the current construction issue in more detail. Thus the chances of becoming "lost in hyperspace" are reduced.
JANUS represents not merely a system building effort but a theoretical framework for application of hypertext to design. Our preliminary evaluation of the JANUS system has demonstrated that hypertext can be used in conjunction with knowledge-based design environments to inform designers effectively, efficiently and without their having to know
1) that they need information, i.e., that their knowledge is inadequate for certain design tasks,
2) that the needed information exists and is in the system, or
3) how to retrieve it.
Within the argumentation system there is a pressing need for authoring to be integrated with browsing. Allowing ad hoc authoring during browsing would enable the designer to annotate the issue base, record decisions on issues and generally personalize the argumentation.
- Instructional and Cognitive Problems
Since hypertext is a new information source, users have not developed skills in navigating, integrating information , and so on. Most novice browsers report discomfort. Many browsers are afraid and don't want to commit the effort. This is partially because they do not have a schema for using hypertext and partly because learners are not inured to the types of information processing stimulated by hypertext. It is a novel form of study. Marchionini (1988) points out that if large amounts of our reading in the future will be by unguided and unconstrained electronic text, new strategies (a hypertext literacy) will be needed. Use with college students has shown that they soon adapt however. Sophisticated browsers develop strategies for searching, comparing information, and solving problems. Effective browsers need to develop hypertext processing strategies in order to effectively use hypertext. How easily can these strategies be taught? Can the system be so well structured and presented as to eliminate the need for any instruction? Will hypertext processing strategies become the information literacy of the 1990's?
The true potential of hypertext for instruction may not be as an information delivery vehicle. Certainly, hypertext is capable of doing that in an unique way. However, its real potential may lie in its capacity as a study aid or cognitive learning tool. A cognitive learning tool is any activity (that may or may not be supported by computers) that fosters or facilitates a deeper or more meaningful level of information processing in learners. Many cognitive learning strategies have been developed to help learners organize, elaborate, and retrieve information. Other computer-based systems have been found to foster these levels of cognitive processing. With other technologies, such as expert systems, we have found that the act of creating the systems engages the learner in a level of analysis and depth of learning that is not elicited by other instructional or learning strategies. Having students create their own hypertexts, especially if they develop hypergraphs/hypermaps , may provide students with the most powerful learning aid yet provided. Research has shown that learning effects are greater for persons involved in developing materials than for those merely using the system(Beeman et al, 1987). So, hypertext may well function best as a study aid that provides multi-dimensional notetaking. The hypertext will not teach the learner. The learner will learn by creating hypertext.
(How often will people exaggerate the creative learning effort that students are willing to commit?? Creating hypertext (as Roy Rada has shown time and time again is something that students will not easily be seduced into doing./ There are no models. However, getting the first draft or prototype out may be easier; and this is only equivalent to having students write essays. Will they then edit these essays and add hypertext links and structure to them --- not as easy as it appears? This will need to be used with care ...)
Instructional Software Design and Production
Harel and Papert (1990) have taken this idea, of having students design software in order to learn, very seriously in a longterm project they have called "constructionism" at the Headlight school. Continuing in the well established Logo tradition of empowering students, they view this work as a way of offering open ended, creative environments for exploring knowledge and learning by reconstructing knowledge (Harel, 1990). This perspective has a natural resonance with the medium of hypertext as viewed and propsoed in this book: facilitating the navigation of knowledge and its restructuring through guided exploration. It also shares with the constructionists the view of Logo not just aas a mathematical or programming tool, but as a multimedia environment that offers users the intrinsic delight of manipulating dramatic and artistic media for self-expression.
LogoWriter
The attractiveness of Logo is considerably enhanced by the capability of combining programs with text, graphics, animaiton, and sound in one multimedia experience. Harel (1988) describes children who take a great delight in designing mulitmedia experiences that others can use to learn. Harel (1990) even provides evidence that these children learn more than a standard Logo group. <<>>
LEGOLogo
In addition to controlling the formation of figures and patterns of color on the screen, LEGOLogo offers that palpable experience of manipualting real objects and controlling shape and movement in three dimensions. The addition of this robotic component along with the pictorial animations that can provide insight into real world control, offers the first real glimpse into future environments of real enactive experience of virtual realities. (see REsnick et al. <<>>)
VideoLogo
Dickinson and Schaffer (1990) describe an extension of LogoWriter that lets students capture images into a computer (IBM PS/2 with PC Eyes digitizer)
FutureLogo