The advent of Internet and multimedia technologies has recently spurred a lot of interest in their applications in education. The LISA project, one of the web sites around world which supplies multimedia distance education, aims to extend human learning by studying various means the Internet can support social interactions and how these different modes of interaction affect the learning effects.

It is true that different forms of social interactions that can be supported by the Internet, the present study focuses on the one-to-one interactions. Four different peer learning models, all allowing one-to-one interactions, were identified. In addition, four distributed learning systems, each of which corresponds to one of the peer learning models, were developed. The following sections briefly describe what these learning models are and how the distributed learning system can be implemented.

Traditional classroom teaching is teacher-oriented, that is, at most of the time, there can only be one-to-many communication and this is mainly from the teacher to the students. A classroom or lecture hall setting normally does not encourage interactions, not even between-peer help or collaboration among students. It is well understood that peers have a private language which is different from that used by teachers. It is also true that students are in a better position to understand the difficulties of their peers. It stands to reason then, that both the private language and the mutual understanding among students are important vehicles for enhancing students' learning, making student collaboration something to be encouraged whenever possible.

The traditional classroom may not be conductive to student collaboration but the computer network provides a unique opportunity for experimenting peer-collaborated learning. True, peer-collaborated learning cannot replace the normal teacher-centred learning situation, but it can be a useful after-class supplement. It is also an important form of distance learning which has fewer teacher-centred activities.

Peer group learning can be in large groups or in small groups. At the outset of this study, we chose small groups learning, in particular, dyads, since this is the simplest. The students' learning task was to write Java programs, a task which would normally require more than an hour's work from each student. Moreover, these peer learning models are synchronous, that is, interactions during the writing of these programmes are performed in real time. This is what makes these distinct from many project-based learning situations in which asynchronous discussion forums are adopted as means for collaboration and the learning tasks are usually so complex that a project may take several months to finish. The following sub-sections describe the four peer learning models employed in this study together with the computer systems that support them.

Reciprocal tutoring is a straightforward peer tutoring protocol. In each round, two students solve two different problems in sequence. They each alternate the roles of student and tutor with the student working on the problem and the tutor observing and advising the student.

Reciprocal tutoring has the advantage over traditional classroom teaching in that there are more peer tutors available than there are teacher tutors since a class has only one teacher while all classmates are potential peer tutors. However, with the pervasive use of the Internet, it is hoped that Reciprocal Tutoring can provide a viable alternative other than intelligent tutoring system In addition, since the one-to-one tutoring done by intelligent tutoring systems which have been proved to be an effective model of learning (Bloom, 1984). An additional advantage in this tutoring method is that the effects of peer collaborative learning on the tutor may also be studied.

One may question the ability of a student, who is still a leaner, to tutor another student. Clearly the student tutor must be more knowledgeable than the fellow student he or she is teaching, making it a necessary condition that there must be a knowledge difference between the student tutor and the student learner. Two measures solve this problem: First, the tutor was provided with the solution as well as detailed explanations annotated in the solution. Second, the student tutors were asked to solve the problems before participating in the reciprocal tutoring. In the future, it is hoped that artificial intelligence techniques can be employed to build an intelligent tutor to help the student tutor.

Reciprocal tutoring has been studied twice by our group previously. The first study explored various versions of reciprocal tutoring model using SuperCard and Mac Lisp in MacQuadra and PowerMac (Chan and Chou, 1997). The machines are connected via AppleTalk and were tested in laboratory. The second allowed students used the system in real life, in their dormitories rather than in a laboratory. The system adopted a client-server architecture on the Internet. The server was implemented by using C++ and Franz Lisp on a Sun Workstation whereas the client was developed by using Borland Delphi on PC (Hwang, W.L. 1996, and Yang, S.S, 1996). The current study differs from the previous ones in two aspects. Firstly, the system was implemented in Java using applets and secondly, the current system is domain independent. It is thus, a generic system and it is simpler than the others.

Figure 2 shows the structure of the Co-working Model, in which learners are required, as a group, to complete tasks given by the computer. This enables the learners to build up their knowledge through communication and collaboration with their partners.

The co-working environment was a common white board (the co-working area) on which students work. This environment forces the learners to find means to share their workload, to collaborate with the partner and to coordinate the whole task, often generating a great deal of discussion and sharing of ideas. It can be seen that communication is an important element of the Co-working model.

The structure of the Working Along Model (Figure 3) is based on the idea that learners are made to feel that they are not alone during the learning process, that there is always a peer working at the other end of the network from whom they could seek advice when they encounter difficulties. This friend is even willing to chat with them. Again, it is hoped that the communication between the peers would help both to widen their scope of probing so as to produce better learning effects.

In this Model, two students are required to solve a problem independently with the exception that either one of dyad can initiate discussion whenever there is a need to do so. When one party initiates a discussion, this step serves two purposes: one party can help the other to solve the problem, but if neither can solve the problem, they are both experiencing the same difficulty and they can share each other's frustration. This is would be the beginning of a process of discussion, mutual support and sharing of knowledge which would be beneficial to both as together they work out the problem-solving strategies.

"Working Along" is a common practice among students. When two or more classmates do homework together, they would initially work independently but when they come across problem, they would naturally discuss them together. The advent of computer network is just the extension of this kind of working relationship to a much larger physical distance. This relationship is beneficial to students both in arousing their motivation to learning and achieving better learning results.

The structure of the Hybrid Model is as shown in Figure 4. This Model is actually a combination of the Co-working and the Working Along Models. As in the Co-working Model, the dyad is required to complete a given task together on a common white board which is a co-working area. The dyad is allowed to choose their working style; they could either work part of the problem on their own and then pool their efforts to arrive at the answer, or they could work together in the co-working area in turn. The Co-working Model does not allow either of these steps, rather, all work written in the co-working area must have the consensus of both parties. The Hybrid Model is thus, not such a strict form of collaborative learning.

The applied domain, i.e., the subject domain in which the problems are to be solved, is Java programming. The 4 systems that enable the above 4 learning models are also implemented by using Java. Since the language has networking and cross-platform features, the implementation of the system is made easier. For the system architecture, the client/server model is employed for all the 4 systems. Detailed structure of these systems can be found in Figure 5.

The design of the collaborative system also takes advantage of the fact that as all 4 learning models are collaborative learning systems which can share the same kernel to handle the collaborative learning tasks. It is only the parts of a collaborative system that are distinctive from the others to be implemented individually (Figure 6). The use of kernel is important since the implementation of future collaborative systems can also be made simpler by including the kernel as a component.

The client of each system serves as the interface between the system and the users and is implemented as Java Applet. A learner only needs is just a WWW browser to link into our homepage to use the learning systems. All 4 learning systems share the same kernel part which includes the following components:

With this kernel in hand, the four learning systems can be easily constructed as follows:

The system provides no assistance to the learner who has no choice but to seek help from the tutor. At all time, there is a part on the tutor's screen showing the tutee's work (right-up corner of Figure 7). Whenever the tutee needs assistance, the tutor immediately knows what the tutee needs and can provide the most appropriate information to the best of his or her knowledge. This is where the learning due to reciprocal tutoring takes place.

Figure 9. Client interface of Co-working System The interface of Co-working System is as shown in Figure 9. A major component of this interface is the co-working area (up side of the screen) where both members work. Anything that is entered by one member is immediately transmitted and displayed at the other end of the network. There is no self-testing area but the learners can communicate with the dialog (left-down area) and the chat areas (right-down-area). The learners can also choose to switch off the problem description, example demo, and on-line information areas by pressing a button at the centre of the screen.

Figure 10 shows the screen display of the Working Along System with the learner's self-working area (left-up corner) and the partner's working area. (right-up). The partner's work is not always updated, a process which takes place only when the learner requests it. When the need arises, the partner at the other end of the network would also be informed. The system is so designed to enable learners to work without being observed.

As all other components of this system are identical to that of the Co-working system, the discussions of these components are omitted.

The screen design of the interface of the Hybrid System is as shown in Figure 11. This system is the combination of Co-working System and Working Along System. Besides the co-working area, there is also a self-testing area where individual learner can test his or her own ideas. The word "Hybrid" reflects this mixture of two systems. Discussions of other components will not be repeated.

Besides the 4 learning systems, there is a need for a general purpose server to handle matters not directly related to learning. The main functions of the server are as follows:

Due to the limitation of resources and the fact that the Working Along System is similar to the TurtleGraph system (Jehng et al., 1994), only the Reciprocal Tutoring System and Hybrid System were tested. Twelve sophomores and juniors of the National Central University of Taiwan were asked to evaluate the two systems. They were first allowed to explore the system freely then they answered 7 questions on their perception of the systems. Results of the data analysis are as follows:

Figure 13. Results of Evaluation of the Reciprocal Tutoring System

Figure 13 shows the results of analysis of the students' perception on the Reciprocal Tutoring System. The subjects found the system helpful although they preferred the passive role of a tutee to the active role of the student tutor. The cause of this may be lack of motivation on the part of the students or that the role of the tutor needs to be made more interesting.

Figure 14 shows the students' evaluation of the Hybrid System. It was found that the students took turns to solve the problems. In general, they found the system helpful, but there was no consensus on whether or not the collaboration was helpful. A majority claimed that they would not like to pair up with learners from another institute. This might point out to the conclusion that generally students cannot benefit from collaboration and that a crucial factor of successful collaboration is the relationship between the partners – the closer the relationship, the better the learning benefits. Further investigation on factors affecting the effects of collaboration would throw more light on this interesting area of study.

.

Figure 14. Results of Evaluation of the Hybrid System