CONCEPTUAL FRAMEWORKS FOR NETWORK LEARNING ENVIRONMENTS: CONSTRUCTING PERSONAL AND SHARED KNOWLEDGE SPACES

Michael J. Jacobson
University of Illinois at Urbana-Champaign and Vanderbilt University

James A. Levin
University of Illinois at Urbana-Champaign

Contact:

International Journal of Educational Telecommunications, 1(4), 367-388.

Permission to publish electronically granted by the publisher the Association for the Advancement of Computing in Education (AACE)


TABLE OF CONTENTS


ABSTRACT

Educational uses of networks are rapidly expanding as the problems of "ease-of-use" and "access" are gradually being solved. However, even as these problems are being solved, the solutions create second-order problems, such as students and teachers becoming overwhelmed with massive amounts of network generated information. In this paper we present conceptual frameworks that characterize some of the unique properties of network learning environments which then can be used to provide systematic guidance to the design of network learning activities and software tools. We illustrate these frameworks by showing how they have helped us design two different tools for educational uses of networks: the Message Assistant and the Learning Resource Server. The utilization of these frameworks to address general issues related to network learning environments is also considered.

INTRODUCTION

Considerable progress is being made in the development of a computer network infrastructure that can serve education. Research is also suggesting ways distributed electronic networks may be employed in instructionally beneficial ways (Bruce & Peyton, 1992; Hunter, 1992; Levin, Riel, Miyake, & Cohen, 1987; Newman, Goldman, Brienne, Jackson, & Magzamen, 1989; Riel & Levin, 1990). Further, improved communication software allows easier access to network-mediated resources such as electronic mail, electronic bulletin board systems, and information servers (e.g., Eudora, Mosaic, Gopher). But the increasing power and ease-of-use of these network software programs is a two-edged sword. On one hand, students and teachers can now easily communicate electronically and access a wide range of information resources. On the other hand, with a huge mega-network like the Internet (10 to 20 million users and over a million servers, and growing daily), teachers and students may be overloaded with hundreds of messages a day or have trouble deciding on which of the millions of Internet servers to look for a particular piece of information. The very richness of these resources imposes not only logistical, but also special cognitive, demands on the user that may unfortunately diminish the learning potential of educational network use (Riel & Levin, 1990; Ruopp, Gal, Drayton, & Pfister, 1993).

A critical area for research involving educational networks is at this juncture of cognition and the network learning environment: the network software tools and the human-computer interface. A new class of tools for network learning environments is needed which preserves the easy utilization of network resources, yet also helps deal with the cognitive complexity associated with distributed network-mediated learning activities. These tools must be based on (a) an understanding of the special characteristics of distributed network learning environments, (b) the situated nature of human cognitive functioning, and (c) ways to support the attainment of substantive educational goals such as learning complex knowledge, problem solving, and independent thinking.

Whereas there has been considerable work dealing with these last two areas (Brown, Collins, & Duguid, 1989; Bruer, 1993; Clancey, 1993; Greeno & Moore, 1993; Lave & Wenger, 1991; Norman, 1993; Suchman, 1987), there has been little consideration of the first. In this paper, we consider a perspective describing some of these "special characteristics" of educational electronic networks, the Distributed Network Learning Framework (Levin & Jacobson, 1992), and the links this framework has with current understandings of human cognitive functioning and new views of classroom learning. First, we review our earlier work on the Distributed Network Learning Framework (Levin & Jacobson, 1992) and then we discuss a Knowledge Spaces conceptual model based on this framework (Jacobson & Levin, 1993; Levin & Jacobson, 1993). We conclude with a discussion of prototype network software tools based on the Distributed Network Learning Framework and the Knowledge Spaces conceptual model.

THE DISTRIBUTED NETWORK LEARNING FRAMEWORK

The Distributed Network Learning Framework (DNLF) consists of three main elements: (a) network mediators and the flow of information and knowledge, (b) expected value of information, and (c) information optimization. We consider these three elements in turn.

Network Mediators and Flow of Information and Knowledge

Fundamental to our view of electronic networks supporting learning activities is that there are a variety of mediators, both human and computer-based, at nodes on the network. These mediators control the flow of information. One important characteristic of electronic networks is the rapid flow of information through the network. Given this rapid movement of network information, decisions must be made at the network nodes about the nature and value of this information. One general result of these decisions is the more gradual flow of organized information or knowledge.

Expected Value of Information

The DNLF is based on the following general principle: Information appearing at each network node is stored locally if the expected value of storing that information is positive. The expected value of information storage is the expected benefit minus the expected cost. For human mediators, the evaluation of expected value can be quite complex and situation specific. In contrast, computer mediators (sometimes called agents) typically determine the expected value more crudely using syntactically oriented algorithms or rules checking for pre-specified text patterns.

The expected value of information is estimated from the probability of needing the information again (i.e., a prediction problem). The simplest approach to prediction is to assume that the future will be like the past, and thus to predict the likelihood of an event occurring in the future will be the same as the occurrence of the event in the past. If the computer-based mediator analyzed its log of past needs for a particular kind of information, it could, under this "the future will be like the past" assumption, predict that information needed frequently in the past will be needed again in the future. The decision could thus be made to locally store information that was frequently accessed in the recent past. Human mediators may also employ other, more sophisticated semantic ways of predicting the future, such as using background knowledge and mental models about a domain to make causal inferences about potential value of information. But in the absence of such knowledge, the "past frequency" rule can be used.

Information Optimization

Related to the expected value of information is another element of the DNLF: Each node in the network attempts to optimize its functioning by storing things that are likely to be used again. Local storage of information thus occurs over time, essentially creating a local "mini-network" or database of information. This local storage of information may also be internalized by the mediator (i.e., resulting in individual learning).

There are several implications of this element of the DNLF. For example, in a learning environment, there would be added value to information that is used repeatedly at a particular node. There would be a tendency to increase the judged probability of needing that information again, and thus the probability would be raised that the mediator at the node would attempt to store or learn the information. This rule also suggests that over time there would be a gradual acquisition of expertise, as the local storage of needed information would also be expected to become increasingly richer and more organized.

One may also regard the optimization occurring at a node as analogous to the knowledge representation and learning processes of accretion, tuning, and restructuring (Rumelhart & Norman, 1978). Initially, there would be the mere accretion or accumulation of information (both in terms of the human representations in memory and the computer-based storage). Over time, this unordered amassing of locally stored information would become unwieldy and would likely exceed the locally available storage. The mediator at that node could then be motivated to develop more differentiated and organized knowledge structures. These new knowledge structures would then serve as the basis from which the mediator would evaluate new network-based information and would be tuned as new information is locally stored and used at that node. Finally, with the continued access of dynamically changing network information and changing learning needs of the mediator, it will likely be necessary to restructure or develop new representational structures. For example, the human mediator might construct new schemas or mental models, while for the network mediator might require new rules or hyperlink configurations to be created.

Illustrating the Distributed Network Learning Framework

To illustrate key components of the DNLF, let us first consider a non-computer example: Why does someone learn something? (See Table 1.) From the perspective of this framework, one reason people learn something is because it is easier than regenerating the information the next time they need it. People do not memorize every telephone number they call. They only do so if they perceive the value of knowing the number by heart to be worth the effort to necessary to commit the phone number to memory. If someone calls Joe's Pizza, she might initially look up the number in the phone book. The next time she needs to call for a pizza, she might need to look it up again. If she calls Joe's Pizza a lot, she may write the number on a piece of paper and stick it next to the telephone. After a while, if she is a real Joe's Pizza fan, she will learn the phone number by heart. From the perspective of the DNLF, the human mediator (the pizza lover) determined that the expected value of the information (phone number of Joe's Pizza) was greater than the cost or effort of storing the phone number externally. The phone number information was therefore initially moved from the phone book to the piece of paper by the phone. Over time, the probability of needing the information again increased (because she was a real Joe's Pizza fan), and she restructured the storage of the information to optimize access to it by finally memorizing the phone number and not needing the paper by the phone.[1]

Table 1
Selected Features of the Distributed Network Learning Framework.

Distributed Network Learning Framework FeatureExample Comments
Information flows through networks based on decisions made by mediators (students/teachers or a computer agent) at each node.Pizza lover (i.e., the mediator) decides initially to look up the number, then to write it down on a note, then to remember it.
Information and knowledge flows toward where the learners need it.The phone number moves from the phone book, to the note, to the pizza lover's memory.
Information appearing at a network node is stored locally if the mediators expect the information to be of value.Pizza lover makes different decisions about locally storing information, first looking it up, then writing it down, and finally learning it. The phone number would not have been stored if the pizza turned out to be bad and thus that piece of information judged to be of little value.
Over time, mediators at nodes would optimize the organization of locally stored information. The pizza lover gradually optimized her ability to access the pizza phone number, as it was initially slow access in the phone book, then faster access with the written note, and fastest when committed to memory.
Network-based information is not a static, fixed "thing," but rather is dynamic, fluid, and changing.
Joe's Pizza may get a new phone number.

KNOWLEDGE SPACES, HYPERTEXT, AND NETWORK LEARNING ENVIRONMENTS

But how are we to convey these distinctive--but unfortunately somewhat complex and abstract--characteristics of network learning environments described by the DNLF to students and teachers? How can this framework be used to inform the design and use of distributed network learning environments? Given research on the value of an appropriate conceptual model to assist users in operating a complex device or computer program (Norman, 1988; Norman & Draper, 1986), we have been developing network software tools based on the DNLF that employ a Knowledge Spaces conceptual model. The Knowledge Spaces conceptual model can function generatively to evoke multiple metaphors and analogies relevant to articulating different conceptual aspects of network learning environments.[2] At the core of this conceptual model is a spatial metaphor that has useful target aspects for describing important characteristics of electronic networks (e.g., "objects can be physically located in different places separated by space" maps to network learning environments where "computers and people are physically located in different places and separated by space"). The Knowledge Spaces conceptual model also suggests other familiar systems that can serve as a coordinated set of analogies for describing network learning environments. For example, "highway networks that connect physically separated places and people for transportation purposes" metaphorically maps to "electronic networks that connect distributed computers for information transmission purposes."

The Knowledge Spaces conceptual model can also be used to articulate abstract epistemic notions about "knowledge structure" and constructivist approaches to learning that are relevant to instructional uses of network learning environments.[3] For example, Wittgenstein (Wittgenstein, 1953) employed the metaphor of knowledge-as-a-landscape in the preface to his Philosophical Investigations, while a "criss-crossing the knowledge landscape" metaphor (inspired by Wittgenstein) has been used for research into a constructivist conception of the nature of learning (Jacobson et al., in press; Jacobson & Spiro, 1995; Spiro, Feltovich, Jacobson, & Coulson, 1992; Spiro, Vispoel, Schmitz, Samarapungavan, & Boerger, 1987). A network learning environment may thus be "viewed" as a knowledge space that can be criss-crossed or explored for different purposes and from different conceptual perspectives, with different learning possibilities afforded by each traversal.

Another aspect of "knowledge spaces" is that they exist along a continuum from personal to shared (Jacobson & Levin, 1993; Levin & Jacobson, 1993) (see Figure 1). Personal Knowledge spaces are constructed for one's individual learning and knowledge utilization purposes (e.g., personal electronic mail messages or a personal "knowledge-base" intended only for a single person's reference or future use). In contrast, shared knowledge spaces are created for information and knowledge dissemination involving larger audiences (e.g., electronic mailing lists, bulletin board news groups, distributed information servers).


Figure 1. The spectrum of knowledge spaces and representative corresponding network resources.

Knowledge Spaces and the DNLF

But how is it that the Knowledge Spaces conceptual model can help convey aspects of the DNLF? This is done in three main ways. First, the spatial nature of this conceptual model presents the movement or "flow" of information over a network as being metaphorically similar to the movement of objects from one location to another in a physically distributed environment. Second, the rapid nature of this movement is evoked by a different metaphor of electrons and electronics. Most students understand that electricity traverses physical space and does so very, very quickly. Third, the "personal and shared" continuum of the Knowledge Spaces model helps evoke the various ways network knowledge may be organized, ranging from one's "personal space" to the more general public "shared spaces." And the various types of mediators on the networks are also suggested. A personal knowledge space may primarily have one's self as the mediator (and one's personal computer-mediator tools or agents) making the determination about the value of network information and whether or not to store it, while the shared knowledge space will have other human and computer-mediators helping to structure the public knowledge space. Our main point here is that the majority of educational network users will be more comfortable with metaphors and analogies that are generatively derived from a core knowledge spaces metaphor than they would be from the more technical DNLF.

Hypertextual Knowledge Spaces

Hypertextual tools are ideally suited for constructing the conceptual interconnectedness that is central to our notion of personal and shared knowledge spaces (Jacobson & Levin, 1993; Levin & Jacobson, 1993).[4] Powerful hypertext and hypermedia technologies have been developed and are now available (Conklin, 1987). A defining structural characteristic of hypertext and hypermedia is the use of hyperlinks between nodes of information. Electronic networks may also be conceptualized in terms of a link and node structure. Not surprisingly, we are now seeing a merging of hypertext/hypermedia and network technologies. In particular, network-based hypertext and hypermedia software tools, such as the World Wide Web servers and browsers (Netscape, Mosaic), are now available for application in network learning environments, and provide flexible and nonlinear access to vast amounts of globally distributed information.

Also, research is beginning to emerge that suggests ways personal hypertext and hypermedia learning environments can help students to learn complex knowledge (Beeman et al., 1988; Jacobson et al., 1994; Jacobson & Spiro, 1995; Jonassen & Wang, 1993; Lehrer, 1993; Shapiro, 1994). Hopefully, future research will document ways in which learning can be promoted through the use of hypertextual shared knowledge spaces as well.[5]

DISTRIBUTED NETWORK LEARNING FRAMEWORK, KNOWLEDGE SPACES, AND NETWORK RESEARCH PROJECTS

Conceiving of activities conducted over network learning environments from the perspectives of the DNLF and Knowledge Spaces suggests that software tools need to provide specific kinds of functionality beyond mere network access or graphical user interfaces. In this section we consider the prescriptive application of the Distributed Network Learning Framework and the Knowledge Spaces conceptual model to the development of tools for network learning environments. We discuss in turn two of our ongoing research projects, the Message Assistant and the Learning Resource Server.

The Message Assistant

The Message Assistant is an electronic mail system designed to assist a network mediator in constructing personal Knowledge Spaces (Jacobson & Levin, 1992; Jacobson & Levin, 1993; Levin & Jacobson, 1993; Levin & Jacobson, 1992). The program offers standard electronic mail options such as message creating, sending, receiving, forwarding, and replying (Figure 2).[6] In addition, the Message Assistant provides two sets of advanced features: hyperlinks between messages and a rule-based mediator for processing of messages. These are the specific features that implement elements of the Knowledge Spaces conceptual model and thus are intended to address characteristics of electronic networks described in the DNLF.



Figure 2. Zero-g World Design Project message first dealing with magnetic shoes from a high school student (top screen) and the initial reply from a Lockhead engineer at the Johnson Space Center (with assigned Message Views displayed).

Hyperlinks and Constructing Personal Knowledge Spaces

As discussed in the first section of the paper, hypertext and hypermedia are technologies well-suited for implementing a Knowledge Spaces metaphor for network learning environments. In a large set of messages generated over a period of time (ranging from weeks to an entire school year) during a distributed network learning activity, there are many messages sent by the participants dealing with different issues, topics, or themes. Hyperlinks may connect related messages on these different issues, topics, or themes. This hyperlinked web of interconnected messages thus would define a personal knowledge space for that network project message set. Furthermore, because of the unique situated contexts of the many participants in the network project (e.g., students of various ages in different classes and parts of the world, teachers, parents, university faculty, scientists, business partners), each one would probably elect to organize or to interconnect the same set of messages in different ways.[7] With a hypertextual Knowledge Spaces tool, multiple dimensions of interconnectedness that exist in the project messages could be specified with different hyperlink sets created by the various users.

The Message Assistant is a prototype of such a hypertextual Knowledge Spaces tool for electronic mail messages. To create a knowledge space with the Message Assistant, we have implemented two classes of hyperlink mechanisms: fixed hyperlinks and variable hyperlinks. [8]

Fixed hyperlinks. Fixed hyperlinks allow quick, nonlinear access between related messages. These "traditional" hyperlinks may be manually created between different messages by the user or automatically by the program when replying to or forwarding a message. Fixed links allow the user to organize and access related messages in a nonlinear manner.

Variable hyperlinks. While fixed hyperlinks are quite useful, unfortunately they must be rigidly specified. For example, if a fixed link is set between two different messages, that link is always there, under all conditions. This means fixed hyperlinks cannot be readily used to represent different conceptual frameworks or information access conditions. For this reason, we have also implemented another class of links in the Message Assistant: variable hypertext links.

To help evoke the "Knowledge Spaces" metaphor in the program interface, we refer to variable hyperlinks as "views" of the locally stored messages.[9] Two default views of the messages are "In View" and "Out View" that correspond to received and sent (or to be sent) messages (Figure 5). The user may then create new views of the messages that share a common topic, theme, issue, or other conceptual/thematic relationship. Each view represents one possible set of hyperlinks between a subset of the messages. Switching to a different view thus re-configures the links between messages, hence the term "variable hyperlinks."

The Message Assistant was used to organize multiple views of the Zero-g World Design Project messages based on a number of topics and themes found in the messages, such as Magnetic Boots, Dishwashers in Space, or Dribbling (see Figure 3). As many messages contained information that dealt with several issues, it was quite common to find such messages in several views (i.e., a heterarchical structure). In addition, the program allows both for the mediator to manually assign a message to a view and for the user-defined rules to automatically assign messages to a view (see below). Thus the Message Assistant allows the user to create multiple views (either manually or automatically) and then to explore or criss-cross the messages from these multiple conceptual perspectives.



Figure 3. In View listing of Zero-g World Design Project messages listed by date with "pop-up" list of message views (top screen) and variable hypertext links of the Magnetic Boots view (bottom screen).

The Message Assistant and User-Defined Message Processing Rules

Unfortunately, the potential exists for educational users of national and international educational networks to become inundated with large amounts of electronic messages. Similar to earlier work on filtering mechanisms for electronic mail (Malone, Grant, & Turbank, 1986), the Message Assistant permits the user to have messages processed through a set of user-defined rules. These rules consist of conditions--Boolean testing for text strings in the different message fields--that can then trigger several possible actions. There are four Rule Actions: Auto Reply, Auto Forward, Priority, and Auto Add Views (see Figure 4).



Figure 4. List of user-defined rules (top screen) and sample user-defined rule.

In the "real world" use of Message Assistant, user-defined rules may automatically check incoming messages and function as a message filter. Messages from a friend or on a specific project might be assigned a high priority, while other messages might be assigned a lower initial priority. The user can also easily ignore the message prioritizations since an overview listing of all messages is available and the user can select any messages to read. But we anticipate that when the user is confronted with a large number of new messages to read, the preprocessing and prioritizations of messages based on the user's own customized set of rules will prove helpful in deciding which messages to read immediately and which to read later.

Furthermore, as students will typically collect a large number of messages in network-based learning activities such as the Zero-g World Design Project, the user-defined rules and the hyperlinking features of the Message Assistant are intended to assist them in constructing their own personal Knowledge Spaces from this information. A possible scenario of use for students in a network learning project would be for an initial set of rules to preprocess and filter incoming messages. Then, over a period of weeks or months, students would evolve different rules to conceptually interconnect messages based on ideas, themes, issues, and so on. They would be thus creating their own personal knowledge spaces. Note that creating such a personal knowledge space out of a large corpus of messages such as the Zero-g World Design project would be much more difficult to accomplish with a more traditional electronic mail program that merely lists messages that have been received or sent. Indeed, it might even be impossible to create a personal knowledge space given the "read-and-delete strategy" that is common among many users because important information about the project would be forgotten and the original messages lost.

Message Assistant and the DNLF

Whereas the previous section considered the Message Assistant from a Knowledge Spaces perspective, there are also several ways in which these features instantiate components of the DNLF. The network mediator and flow of information and knowledge component holds that probabilistic evaluations of network information are made by a mediator at a particular node. The Message Assistant rules function as a computer-based mediator that assists a human mediator in making an initial determination of the expected value of the information. The message rules form a user-defined expert system which is incrementally specified and tested over time and thus gradually increases its expertise with use. Such an expert system serves as a computer-mediator that takes over repetitive or low-level evaluations of the expected value of certain types of network information and then automatically initiate actions which filter, route, or store that information. Information that does not match any pre-specified rules is passed on to the human mediator to be evaluated. If the new information is regarded as being of value, then the human mediator will operate on that information.

The variable hyperlinks feature also helps to instantiate the DNLF principle of optimization (i.e., network nodes attempt to optimize their functioning by locally storing potentially useful information). The message Views provide the user with a mechanism to create a personal knowledge space that is also flexible in terms of organizing and accessing the information in the messages. Over time as the human mediator works with the dynamic and changing learning environment, the specification of message views and rules would be expected to progress from gradual accretion, to the tuning use of a body of accumulated views and rules, to the restructuring of views and rules.

Learning Resource Server

Tools for creating hyperlinks are ideally suited for structuring multiple organizing frameworks for any given set of information. This type of functionality is especially important when a wide range of people need to flexibly access and use information, as is the case with network-based information servers. However, merely providing flexible access to information with hyperlinks is not sufficient to ensure substantive learning will occur (Jacobson, 1994; Jacobson et al., in press; Jacobson & Spiro, 1995). As we noted above, research is beginning to identify the theoretical and design characteristics of effective hypertextual learning environments, such as: active, nonlinear student exploration of hyperlinked information nodes, student creation of new hyperlinks in existing materials, explicit depiction of important interrelationships between surface and structural knowledge components across multiple case examples, and student authorship of hyperlinked materials (Beeman et al., 1988; Jacobson et al., 1994; Jacobson et al., in press; Jacobson & Spiro, 1995; Jonassen & Wang, 1993; Lehrer, 1993; Shapiro, 1994).

We are investigating the application of research findings such as these to the development of a network-based Learning Resource Server. The server, currently under development, contains a large hyperlinked knowledge-base of education projects, curriculum units, research reports and papers, and links to other network servers with documents relevant to education. Figure 5 shows a sample screen from the UIUC Learning Resource Server. Overall, the UIUC Learning Resource Server is intended to provide a shared knowledge space that can be utilized by a wide range of learners and researchers. We are also exploring how tools for constructing hyperlinked personal knowledge spaces, such as the Message Assistant, can be used by students in conjunction with the shared knowledge space of the Learning Resource Server.


Figure 5. Home page of the University of Illinois at Urbana-Champaign Learning Resource Server (URL http://www.ed.uiuc.edu/lrs/).

CONCLUSION

In this paper, we have discussed conceptual frameworks for systematically developing software tools for network-based learning environments that can help students and teachers construct personal and shared Knowledge Spaces. We have described a software tool based on these conceptual frameworks, the Message Assistant, that allows a user to construct a personal knowledge space. We have also sketched out our initial efforts at applying similar concepts and tools to the design of Learning Resource Servers that can be used to construct hyperlinked shared Knowledge Spaces.

For network learning environments to have a substantial and positive impact on education, "ease-of-use" and "universal access" are not enough. Students and teachers need conceptual frameworks to help organize their activity, they need tools that are consistent with such frameworks, and they need mediators to enable the activity. This paper describes some theoretical and research steps toward these goals.

REFERENCES

Beeman, W. O., Anderson, K. T., Bader, G., Larkin, J., McClard, A. P., McQuillan, P. J., & Shields, M. (1988). Intermedia: A case study of innovation in higher education (Final report to the Annenberg/CPB Project). Providence, RI: Brown University, Office of Program Analysis, Institute for Research in Information and Scholarship.

Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the culture of learning. Educational Researcher, 18(1), 32-42.

Bruce, B. C., & Peyton, J. K. (1992). A situated evaluation of computer networking to teach writing (Technical Report No. 565). Urbana: University of Illinois, Center for the Study of Reading.

Bruer, J. T. (1993). Schools for thought: A science of learning in the classroom. Cambridge, MA: The MIT Press.

Clancey, W. J. (1993). Situated action: A neuropsychological interpretation response to Vera and Simon. Cognitive Science, 17, 87-116.

Conklin, J. (1987). Hypertext: An introduction and survey. IEEE Computer, 20(9), 17-41.

Edelman, G. (1992). Bright air, brilliant fire: On the matter of the mind. New York: Basic Books.

Feltovich, P. J., Spiro, R. J., & Coulson, R. L. (1989). The nature of conceptual understanding in biomedicine: The deep structure of complex ideas and the development of misconceptions. In Evans, D., & Patel, V. (Eds.), The cognitive sciences in medicine (pp. 113-172). Cambridge, MA: MIT Press (Bradford Books).

Greeno, J. G., & Moore, J. L. (1993). Situativity and symbols: Response to Vera and Simon. Cognitive Science, 17, 49-59.

Hunter, B. (1992). Linking for learning: Computer-and-communications network support for nationwide innovation in education. Journal of Science Education and Technology, 1(1), 23-34.

Jacobson, M. J. (1994). Issues in hypertext and hypermedia research: Toward a framework for linking theory-to-design. Journal of Educational Multimedia and Hypermedia, 3(2), 141-154.

Jacobson, M. J., Maouri, C. , Mishra, P., & Kolar, C. (in press). Learning with hypertext learning environments: Theory, design, and research. Journal of Educational Multimedia and Hypermedia.

Jacobson, M. J., & Levin, J. A. (1992). A rule-based electronic mail processor for collaborative electronic learning environments. Paper presented at the annual meeting of the American Educational Research Association, San Francisco, CA.

Jacobson, M. J., & Levin, J. A. (1993, December). Hypertext and network-based learning environments: Technology for the construction of personal and shared knowledge spaces. Paper presented at the International Conference on Computers in Education: Applications of Intelligent Computer Technologies, Taipai, Taiwan.

Jacobson, M. J., & Levin, J. A. (1993). Network learning environments and hypertext: Constructing personal and shared knowledge spaces. In Foster, D., & Jolly, D. V. (Eds.), Proceedings of Tel-Ed `93 (pp. 190-197). Dallas, Texas: International Society for Technology in Education.

Jacobson, M. J., & Spiro, R. J. (1995). Hypertext learning environments, cognitive flexibility, and the transfer of complex knowledge: An empirical investigation. Journal of Educational Computing Research, 12(5), 301-333.

Jonassen, D. H., & Wang, S. (1993). Acquiring structural knowledge from semantically structured hypertext. Journal of Computer-based Instruction, 20(1), 1-8.

Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge: Cambridge University Press.

Lehrer, R. (1993). Authors of knowledge: Patterns of hypermedia design. In Lajoie, S. P. & Derry, S. J. (Eds.), Computers as cognitive tools (pp. 197-227). Hillsdale, NJ: Lawrence Erlbaum.

Levin, J., & Jacobson, M. (1993). Educational electronic networks and hypertext: Constructing personal and shared knowledge spaces. Paper presented at the annual meeting of the American Educational Research Association Conference, Atlanta, Georgia.

Levin, J. A., & Jacobson, M. J. (1992). Towards a distributed network learning framework: Theory and technology to support educational electronic learning environments. Proceedings of the Fourteenth Annual Conference of the Cognitive Science Society (pp. 927-932). Hillsdale, NJ: Erlbaum.

Levin, J. A., Riel, M., Miyake, N., & Cohen, M. (1987). Education on the electronic frontier: Teleapprentices in globally distributed educational contexts. Contemporary Educational Psychology, 12, 254-260.

Malone, T. W., Grant, K. R., & Turbank, F. A. (1986). The information lens: An intelligent system for information sharing in organizations (CISC WP Report No. 133). Boston, MA: Center for Information Systems Research.

Newman, D., Goldman, S. V., Brienne, Jackson, I., & Magzamen, S. (1989). Peer collaboration in computer-mediated science investigations. Journal of Educational Computing Research, 5(2), 151-166.

Norman, D. (1993). Cognition in the head and in the world: An introduction to the special issue on situated action. Cognitive Science, 17(1), 1-6.

Norman, D. A. (1988). The design of everyday things. Reading, MA: Addison-Wesley.

Norman, D. A., & Draper, S. W. (Eds.). (1986). User centered system design: New perspectives on human-computer interaction. Hillsdale, NJ: Lawrence Erlbaum.

Riel, M., & Levin, J. A. (1990). Building electronic communities: Success and failure in computer networking. Instructional Science, 19, 145-169.

Rumelhart, D. E., & Norman, D. A. (1978). Accretion, tuning and restructuring: Three modes of learning. In Cotton, J. W. & Klatzky, R. L. (Eds.), Semantic factors in cognition (pp. 37-53). Hillsdale, NJ: Lawrence Erlbaum.

Ruopp, R., Gal, S., Drayton, B., & Pfister, M. (1993). LabNet: Toward a community of practice. Hillsdale, NJ: Lawrence Erlbaum.

Shapiro, A. M. (1994). Complex concept acquisition and the representation of knowledge: A study in hypermedia-aided instruction. Paper submitted for publication.

Spiro, R. J, Feltovich, P. J., Coulson, R. L., & Anderson, D. K. (1989). Multiple analogies for complex concepts: Antidotes for analogy-induced misconception in advanced knowledge acquisition. In Vosniadou, S., & Ortony, A. (Eds.), Similarity and analogical reasoning (pp. 498-531). Cambridge, MA: Cambridge University Press.

Spiro, R. J., Feltovich, P. J., Jacobson, M. J., & Coulson, R. L. (1992). Cognitive flexibility, constructivism, and hypertext: Random access instruction for advanced knowledge acquisition in ill-structured domains. In Duffy, T. M., & Jonassen, D. H. (Eds.), Constructivism and the technology of instruction: A conversation (pp. 57-75). Hillsdale, NJ: Lawrence Erlbaum.

Spiro, R. J., Vispoel, W. P., Schmitz, J. G., Samarapungavan, A., & Boerger, A. E. (1987). Knowledge acquisition for application: Cognitive flexibility and transfer in complex content domains. In Britton, B. K., & Glynn, S. M. (Eds.), Executive control processes in reading (pp. 177-199). Hillsdale, NJ: Lawrence Erlbaum.

Suchman, L. (1987). Plans and situated actions: The problem of human-machine communication. Cambridge: Cambridge University Press.

Wittgenstein, L. (1953). Philosophical investigations. New York: Macmillan.

ACKNOWLEDGMENTS

This material is based upon work supported by the United States National Science Foundation under Grant No. RED-9253423. The United States government has certain rights in this material. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. This work was supported by a grant of equipment from Apple Computer, Inc. Portions of this paper have been revised from earlier conference presentation papers on this research. The authors acknowledge the contributions of Youngcook Jun and Yasuhiro Uno for their programming work on the Message Assistant. We also thank Matthew Stuve, Pia Bombardier, and Evangeline Secaras for their assistance on establishing the University of Illinois College of Education Learning Resource Server.

NOTES