Scaffolding Group Learning in a Collaborative Networked Environment Amy S. Wu University of Michigan Ann Arbor, MI 48109
[email protected]
Rob Farrell, Mark K. Singley IBM T.J. Watson Research Center Yorktown Heights, NY 10598 {robfarr, ksingley}@us.ibm.com
ABSTRACT Scaffolding students in a collaborative networked learning environment requires different instructional methods than in a traditional home or classroom setting. The goal of this research is to understand computer-mediated collaboration in an instructional setting in order to create an effective computer-mediated collaboration tool. We identify ways to support collaboration by examining the interaction and strategies employed by a peer tutor and teacher and between peers working in our collaborative learning environment. We found that supporting collaboration in an electronic setting requires diagnosing impasses, facilitating problem-solving interaction, and suggesting ways to divide the problem into sub-tasks. Keywords computer-mediated collaboration, tutoring, learning environments
INTRODUCTION Proponents of collaborative learning claim that students in cooperative teams achieve higher levels of thought and retain information longer than students who work quietly as individuals (Webb, 1995). Students learning effectively in groups encourage each other to ask questions, articulate their thoughts, explain and justify their opinions, and elaborate and reflect upon their knowledge. Yet, the field of CSCL (Computer-Supported Collaborative Learning) continues to struggle to identify how best to support collaborative learning online as new communication technologies allow the learner to network with other learners (Berge, 1997) and as CSCL becomes an established research paradigm for electronic learning environments (Koschmann, 1996). How does a collaborative networked learning environment impact the types of tutoring needed to effectively scaffold students? We report on a study that compares peer-mediated instruction and teacher-mediated instruction to inform scaffolding in a collaborative networked learning environment. The goal of this research is to investigate how peers and teachers support group learning in a networked environment so in the future, we can use this model to inform the design of electronic scaffolds for our learning environment.
THEORETICAL RATIONALE The following sections provide a foundation and rationale for our work. A brief discussion of collaborative learning and technology-based learning environments will establish a framework for our study. Collaborative Learning Many educators and learning researchers have found promise in Vygotsky’s (1978) social constructivist ideas about learning in a social setting. His notion of the zone of proximal development (ZPD) posits that children’s mental development can be positively influenced by the assistance of an adult or more capable peer. The core concept of constructivism is that understanding and knowledge come from our interactions with the environment (Savery & Duffy, 1996). Studies suggest that communicating ideas in a group setting encourages self-explanation and justification, both of significant instructional value (Rogoff, 1990; Webb, 1991). Different groupings of learners with or without tutors have been shown to be effective in improving student achievement. A tutor may be a teacher or a more capable peer, both whom Vygotsky (1978) believed could guide a learner to reach his or her potential level of intellectual development. Bloom (1984) demonstrated that tutoring
raises students’ performance by .40 standard deviations with peer tutors and 2.0 standard deviations with experienced tutors. What strategies do experienced tutors use to facilitate learning? Graesser, Person and Magliano (1995) documented the techniques used by tutors during the evolution to a problem-solving solution: pumping to expose student knowledge; prompting to supply the student with context to fill in missing information; and splicing to provide the student with correct information when they make a mistake. Other tutoring strategies found by Graesser et. al. include hinting, summarizing, elaborating, questioning, conducting assessment, and providing immediate feedback. Similarly, in a study of expert human tutors, Lepper, Woolverton, Mumme, and Gurtner (1993) found that experienced tutors virtually never provided the answer to a problem. Instead, they responded to errors more indirectly with hinting and prompting strategies. As Bloom (1984) showed, peers can also mediate each other’s learning. Peer tutoring is an arrangement in which a more knowledgeable peer assists a less knowledgeable one. McCarthey and McMahon (1992) found that in a peertutoring situation, the discourse tends to be unidirectional, from the more knowledgeable student to the less knowledgeable student. Jenkins (1974) compared one-to-one peer tutoring with conventional teacher-led, small group lecture-style instruction of students with learning disabilities and reported that the tutoring procedures were significantly more effective than the teacher-mediated method. Few studies have documented instructional strategies of peer tutors during unstructured interaction, excluding studies of structured approaches such as Reciprocal Teaching (Palincsar & Brown, 1984). Same-level peers, however, may not collaborate effectively to mediate each other’s learning. Nelson-LeGall (1992) suggests that with schooling, students perceive that competition and independent performance are increasingly normative, which may account for their failure to seek help from their peers and even from their teachers. Other studies (Van der Miej, 1988; Newman & Goldin, 1990) found that students seeking help in the traditional classroom are concerned about the perceived costs of seeking help, such as uncooperative reactions and appearing dependent or incompetent. However, individuals need to exchange ideas, information, and perspectives for interaction to promote learning. For example, King (1989) found a positive relationship between verbal interaction and children’s problem-solving success. In comparing the effectiveness of working with experts or peers, Salomon and Perkins (1998) note an interesting difference between expert tutoring and peer problem solving. Whereas tutors aim to facilitate the learning of the student(s), peers working together often aim just to accomplish the task. As a result, the individual learner often gets more of a chance to participate actively in the problem-solving effort when interacting with an expert tutor than with same-level peers. Similarly, Vedder (1985) found that students working in groups appear to be more focused on finding the right answer than in mediating each other’s learning. Therefore, in many cases, an expert tutor needs to support peer problem-solving. Why scaffold collaborative learning? Despite studies (e.g. Webb, 1991) showing that social learning situations correlate with a wide range of positive outcomes – such as improved learning, increased productivity, and higher motivation – collaborative learning does not work for all learners. According to Brown and Palincsar (1989), “social interactions do not always create new learning; peer interactions vary enormously.” (p. 397). Several researchers (King, 1994; Webb, 1995) have found that it is the nature of peer interactions that is perhaps the most critical factor mediating individual student achievement. The human tutoring literature shows that expert tutors can effectively support a single student or group of students, and that peers (both more capable peers and same-level peers) can mediate each other’s learning.. However, these studies of tutoring are based on individual tutoring situations or face-to-face group instruction, which may not necessarily apply to networked learning. The next section explores technology-based learning environments that can support collaborative learning. Technology-based learning environments Technology has taken on new roles in learning, in ideal situations, providing students with the opportunity to interact with responsive, dynamic environments that support learning. These computer-based interactive learning environments have traditionally been designed to support a single student’s needs. Interactive learning environments (ILEs) are commonly grounded in constructivist theories on learning, which emphasize that knowledge is something that a student constructs based on his or her prior knowledge and experience. Learning needs to be active and problems should be embedded in realistic and relevant contexts (Knuth & Cunningham, 1993). ILEs are designed for a single student but may have a computer tutor to assist in learning. Computer tutors in interactive learning environments, also known as Intelligent Tutoring Systems (ITS), can diagnose the conceptual
difficulties and misunderstandings of any student (Anderson, Corbett, Koedinger et al., 1995). Model tracing is one approach that is used to monitor student progress through a solution path (Anderson, Boyle, Corbett et al., 1990). This tracing is done by matching student actions to those the model might generate. Brown and VanLehn (1988) found that learning occurs when students encounter problems, at which point they need help to repair their incorrect procedures. CSCL environments, in addition to supporting learning of subject matter, also support students working with others. These systems have the potential to enhance the effectiveness of peer interactions, by coaching peers as they work on problems and critique other students’ solutions. A human or computer tutor can help structure the interaction, give advice when needed, and promote deepening of understanding. CSCL environments can support a single student working with a human tutor (expert or peer) or with same-level peers or with both an expert tutor and peers, corresponding to the collaborative learning arrangements described in the previous section. This paper explores various types of collaborative learning systems to understand how they differ from traditional ILEs and why it is worthwhile to study human tutoring in a CSCL environment. In CSCL environments, in which students work together at a distance or at the same computer, the task of learning requires additional tools for students to work collaboratively with others. Various types of systems exist to support collaborative learning (Jermann, Soller & Muehlenbrock, 2001). Systems that reflect the actions of each user, termed mirroring systems, make the students or teachers aware of other participants’ actions. For example, in the Learning Through Collaborative Visualization (CoVis) project, an application called the Collaboratory Notebook (Edelson & O'Neill, 1994) supports learners through the scientific inquiry process and allows them to share their work with others. Such shared spaces can be achieved through desktop video conferencing and remote screensharing, which allows individuals at one end to view the contents of their partners’ computer screen and to control the remote computer with their own keyboard and mouse. Systems that monitor the state of interaction, called metacognitive systems, provide participants with information about exchange patterns. CSILE (Scardamalia, Bereiter, McLean et al., 1989) scaffolded collaboration by asking students to select a category for their comment(s) during participation in group discussion. The Reflective Learner (Guzdial, Hmelo, Hubscher et al., 1998) requires that students abstract high level principles from observations of the collaboration and relate or project them to past and future experiences. These systems support the state of interaction by requiring participants to reflect on the collaborative experience. Finally, coaching systems guide the collaborators by recommending actions participants might take to improve their interaction. The coach (human or computer) is responsible for guiding the students toward effective collaboration and learning. The facilitator must address collaboration issues such as distribution of roles among students, equality of participation, and reaching a common understanding (Teasley & Roschelle, 1993). For example, GRACILE (Ayala & Yano, 1998) contains computer agents that assess the progress of individual learners and propose new tasks based on the needs of the group to help students learn Japanese. From studies of an ITS called Sherlock (Lajoie & Lesgold, 1989; Lesgold, Lajoie, Bunzo et al., 1992), Katz and Lesgold (1993) believe that the tutor has three main roles to play in collaborative learning situations: 1.
provide advice on demand
2.
provide quality control over collaborative activities
3.
manage peer collaboration
Our study also investigates the role of a tutor in a computer-support collaborative learning environment, but we conduct an empirical study within human tutors to answer this question. The three types of CSCL environments – mirroring, metacognitive, and coaching – provide supportive spaces for students to collaborate. However, none of these environments discussed based their design of support for collaboration on studies of human tutors and facilitators. Our study is designed to understand the effective strategies of human tutors in a CSCL environment, in which networked computers link the participants. What is the nature of peer and expert tutoring? How does such an environment impact the types of tutoring needed to effectively scaffold students? The goal of our research is to determine how best to support learning in a CSCL environment. Our longterm goal is to use the results of this study to inform our design of electronic scaffolds for a group of students working together on tasks requiring conceptual understanding.
METHOD We conducted a small study that examined the role of a teacher or more experienced peer with one or two other students in a challenging problem-solving situation. We chose algebra as our subject domain because algebra provides concrete, well-defined problems on which students can collaborate. Our existing system, ALGEBRA JAM (Singley, Fairweather & Swerling, 1999), supported collaborative problem-solving and was an appropriate environment for exploring various collaborative arrangements. Collaboration in ALGEBRA JAM ALGEBRA JAM is a CSCL environment that supports teams of students as they collaborate synchronously and remotely to solve situated, multi-step problems involving algebraic modeling. Students are provided with resources containing information they need to solve an algebra word problem and given various tools to support collaboration. These tools include a blackboard, a scratchpad, and a topical chat utility. Each will be discussed briefly here. The blackboard (shown in the top right of Figure 1) supports the participants as they create a “goal tree”, share ideas about how to attack a problem, and monitor team progress. They can post the variables they need to compute as nodes in the goal tree. The top of the tree contains the final goal. For example, in the problem shown, the final goal is to determine the time to start mowing Mr. Sweeney’s lawn. To the left of the tree is a description of the selected node, including the name of the student who posted it and the person who last modified it. All participants in the problem-solving session can create a node in the goal tree or change the contents of an existing node. In the example, one student may realize that to find the time to start, they would need to know the finish time and the time it would take to mow the lawn, two variables to post in the tree by clicking “New” and selecting those variables from a list. Any changes in the blackboard are reflected synchronously to other participants in the session; the students cannot make their work private. To aid in collaboration, a face of the user(s) with whom a student is collaborating is displayed in a panel to the right of the blackboard. As Gutwin, Greenberg, and Roseman (1996) note, features that support the awareness of others in the environment facilitate collaboration.
Figure 1: ALGEBRA JAM interface The tabbed workspace (shown in the lower right of Figure 1) is an area where students work with the various information resources required to solve the problem. The resources are dragged from the bookshelf (shown in the upper left of Figure 1). Like the blackboard, when one participant performs an action in the tabbed workspace area, that action is reflected synchronously to all other participants. In this regard, Jam is a mirroring system, which reflects the actions of each user. This is intended to encourage collaboration in that students can work together to write equations, perform calculations, model solutions, and evaluate each other’s work. Rather than provide a generic chat facility, ALGEBRA JAM allows learners to participate in threaded discussions tied spatially to the work products and interface objects with which they were working. Learners communicate with one
another by typing a message while pointing to a particular object on the screen. They click on “Chat” above the face of other participants and a balloon opens in which they type. Users can place the chat balloon anywhere on the screen to focus their conversation or chat about a particular action or object. To further support collaboration, users must identify the type of comment they are contributing. CaMILE (Guzdial, Carlson, & Turns, 1995), a Web-based collaboration tool for use by students to encourage learning, also includes prompts for students to identify the kind of notes they are contributing to the collaboration and suggests things to say in those notes. When a user indicates that they want to create a new note, they are prompted to identify the type of their note. The goal is to encourage metacognition in students. Similarly, ALGEBRA JAM contains buttons called ‘collabicons’ which students and teachers use to categorize, or assign a type, to their message. The five collabicons include “Thumbs Up”, “Thumbs Down”, “Here’s Help”, “Here’s An Idea” and “Help Me.” This feature provides metacognitive support. In our description of ALGEBRA JAM , we have highlighted features that support collaboration, which illustrate its effectiveness as a mirroring and metacognitive system. However, as for coaching, no features exist yet to support collaboration between participants. How does a CSCL environment impact the types of support needed to effectively scaffold collaborative learning? Participants The participants were eighteen 8th-12th graders from summer school programs in the metropolitan New York area. Six of the students were advanced math students, in higher-level math classes beyond algebra such as trigonometry, while the other twelve were currently enrolled in or retaking an algebra class. The students who participated in the same groups knew each other. The teacher was a high-school math teacher with 22 years of teaching experience. The peer tutor was a student who also participated in our study as a (non-tutor) student and agreed to return in the role of a tutor in subsequent experiences. Design of Study We placed participants in one of four experimental groups as arranged in Figure 2. We observed three sets of participants in each condition.
Teacher Student(s)
none one two(no roles) two(roles)
one Expert Tutoring
Peer Collaboration
Collaboration Plus
Peer Tutoring
Figure 2: Experimental Groups
The “Expert Tutoring” condition involved one student working with one teacher. This condition corresponds to the standard one-on-one tutoring situation that has been studied extensively (Bloom, 1984; Anderson & Skwarecki, 1986). We expected that students in this condition would regularly consult the teacher for advice and suggestions. The “Peer Collaboration” condition involved two students working together without role assignment (i.e. neither student was told to tutor the other). We considered this the pure collaboration condition in which the pair would work together more or less as equals to solve the problem. We expected that when two students were present, they would work closely, collaborating on most decisions, since they were provided tools to support collaboration and a design that encouraged working together. “Collaboration Plus” involved two students working with a teacher. We were interested in how adding a teacher to the learning environment influences collaboration between peers. We hypothesized that students would approach the teacher more than they would consult one another for help. “Peer Tutoring” involved two students working together, in which one played the role of a tutor. The peer tutor had been a student who participated in the “Collaboration Plus” condition, and then placed with a student who had never used ALGEBRA JAM before. The peer tutor had been given instruction on the problems before assuming the tutor role. We suspected that the peer tutor would be approached in a manner similar to how the teacher was approached.
Procedure When students initially arrived to participate in the study, we took digital photos of them to load into the system. Prior to using the system, the participants were instructed on its features as we demonstrated a scenario between two users. The students in the collaboration conditions were shown how to chat with each other and reminded that any work they did in the blackboard or workspace would be reflected in the other student’s interface. The students working together were not directed on how or when to collaborate and were merely told that they would be working together using computers in different rooms. The students in the same groups knew each other previously. In the “Expert Tutoring” and “Collaboration Plus” conditions, the teacher was introduced as a resource if the student(s) encountered difficulties. The teacher was not provided direction on how frequently to provide assistance to the student(s). The teacher, placed in a separate room than each student, could see the events that each of the students saw on their computer. Each participant was videotaped as they used the software on the computer. The ALGEBRA JAM environment was augmented to automatically create logs of the chats and the activity in the blackboard and workspace. Each group of participants was given a set of three problems to solve in one hour using the system. Each problem involved a landscaping scenario in which students had to solve for variables such as mowing rate or the time to complete a job. The same resources (maps, work logs, a white pages directory) were available on the bookshelf throughout the session, so the student(s) needed to decide which subset would be useful for each particular problem. Analysis No measures were taken to assess learning because of the short timeframe in which they used the system and the design of the open-ended algebra problems. We focused on the nature of the collaboration and did not assess cognitive gains, although we did record how many problems each student or group attempted and solved. From the videotapes and the logs of the tutoring sessions, we analyzed the manner in which the human tutors offered assistance and responded to student successes and struggles, and the attempts the tutors made to articulate and instruct students in the principles underlying the problem solution. We also coded the participants’ actions and decisions so we could identify general strategies and techniques employed by peer tutors and teachers when students encountered difficulties. We identified types of assistance and feedback offered by a tutor, calculated the number of times a tutor mediated the problem-solving effort, and determined the success of each group by calculating problems solved versus problems attempted.
FINDINGS We present our findings at two levels. First, we examine the collaboration, or the flow of information, that occurred between participants in ALGEBRA JAM . We begin by looking at how often learning is mediated and by whom. Then, we discuss how the participants mediated each other’s learning. Modeling information flow Information flow is defined as the sharing of meaningful, relevant information that assists in completing the task. We have modeled the information flow between participants in each condition to summarize our findings in Figure 3. The data to support these diagrams is discussed in the following sections. We had expected to see arrows between the two students in the collaboration conditions, but no useful information was exchanged between peers.
Teacher none
one S
Student(s)
one
T
Expert Tutoring two(no roles)
S1
S2
Peer Collaboration two(roles)
S
S1
T
S2
C Collaboration Plus
PT
Peer Tutoring
Figure 3: Observed information flow
In the “Tutor mediation” tables in this section (Tables 1, 2, and 3), we list the number of problems which the student(s) attempted, the total number of times the teacher intervened during the session, the percentage of interventions solicited by the student, and the percentage of interventions initiated by the teacher during that session. “Peer Collaboration” does not have a table since no tutor participated in that condition. Expert Tutoring In “Expert Tutoring”, the students regularly consulted the teacher for advice and suggestions. In fact, it seemed that students had the perception that they were collaborating with the teacher to solve the problem. There was a constant flow of conversation between the tutor and the student. Most of the information was flowing from the teacher to the student, as illustrated in Passage 1. Tutor: Good, now how many more minutes is the first time besides 120? Timmy: So is it 120.15? Tutor: 120 minutes + 15 minutes = Timmy: 135
Passage 1: Information directed from teacher to student In the “Expert Tutoring” condition, most of the interaction involved the problem-solving process. On average, the student solicited approximately 45% of the teacher interventions in this condition; 55% were initiated by the teacher. Only one of the three students progressed past the first problem. Table 1 shows that his session was the only one in which the percentage of solicited interventions was greater than the percentage of tutor-initiated interventions. Expert Tutoring Group 1 Group 2 Group 3 Average
Problems Attempted 1 2 1 1.3
Tutor Interactions 13 9 9 10.3
Student-Solicited Interactions n(%) 6(46.1) 5(55.5) 3(33.3) 4.7(45.0)
Tutor-Initiated Interactions n(%) 7(53.8) 4(44.4) 6(66.7) 5.7(55.0)
Table 1: Tutor mediation for “Expert Tutoring” Peer Collaboration In the “Peer Collaboration” condition, long periods of time passed without any chatting. Students would do their own work, as evidenced by Passage 2 from our chat transcripts. The students are working on a problem called “Who Should We Hire?” in which they need to compare the mowing rate of two job candidates, Phillip and Tom. To do so, they open the resources “Phillip’s Work Log” and “Tom’s Work Log” in the workspace (see Figure 1). Because the workspace is tethered, they need to collaborate on the same resource to make any progress on the problem. For example, one student cannot be computing Phillip’s rate while the other student is computing Tom’s rate; they must work together because their actions are reflected on each other’s workspace. Two same-level peers did not mediate each other’s learning, which explains why the diagram in Figure 3 did not have any arrows representing the information flow between two students in the Peer Collaboration condition. Ashley: Stay in one place. Ashley: I can’t finish the problem. ... Lanice: I think I have solved the problem… Ashley: How did you get that? Ashley: Good job
Passage 2: Lack of peer mediation Passage 2 is the end of a chat transcript in which Lanice and Ashley worked independently to solve the problem. After Lanice is done, Ashley inquires about Lanice’s efforts but just ends the session by congratulating her. Although the two students are friends, they are not gaining anything by having a peer present. Collaboration Plus In “Collaboration Plus”, where a teacher and a same-level peer are present for consultation, we found that neither the teacher nor the peer were asked directly for help. Sometimes, students would lead each other down the wrong path and the teacher would voluntarily intervene. Remarks are often undirected. The students did not share useful
information, although they did interact more with a teacher present than they did in the “Peer Collaboration” condition without a teacher. They usually only interacted to encourage each other or put each other down, as in Passage 3. Tutor: You have total acres and total hours. How do you get the rate? Aaron: Ohhhhh my bad. Brandon: Yes it is your bad. Aaron: Shut up. I didn’t hear you come up with some answer, Mr. I-know-it-all.
Passage 3: Peer interaction in “Collaboration Plus” With multiple students, the teacher, instead of being solicited by the learner(s) as in “Expert Tutoring”, initiates more of the mediation. Table 2 shows that a large percentage (90% average) of the tutor mediation in “Collaboration Plus” was tutor-initiated, compared to Table 1 in “Expert Tutoring” which showed tutor-initiated interactions averaging 55%. One group of students was able to progress through all three problems. This group, Group 2, also had the lowest number of tutor interactions. Collaboration Plus Group 1 Group 2 Group 3 Average
Problems Attempted 1 3 1 1.3
Tutor Interactions 17 10 15 14.0
Student-Solicited Interactions n(%) 1(5.9) 1(10.0) 2(13.3) 1.3(9.7)
Tutor-Initiated Interactions n(%) 16(94.1) 9(90.0) 13(86.7) 12.7(90.3)
Table 2: Tutor mediation for “Collaboration Plus” Peer Tutoring In “Peer Tutoring”, the peer tutor was consulted in a similar fashion to the teacher in “Expert Tutoring”. No fewer questions were asked than in the condition where a teacher was present. The peer tutor seemed to be an adequate substitute for the teacher in terms of answering questions and directing students. The following transcript passage, Passage 4, is an example of effective guidance by a peer-tutor. Peer Tutor: No…what are u trying to find in the “Phillip” book? Tremaine: The amount of acres that he mows per hour Peer Tutor: Yes…his average rate of lawn mowing. How does one find an average? Tremaine: Divide
Passage 4:Effective peer tutoring Students interact with a peer tutor in a similar manner as they interact with a teacher, both in asking questions and soliciting help. Information was directed from the tutor to the student, just as we saw in the “Expert Tutoring” condition. As shown in Table 3, our data set demonstrates that a peer, with a lower average number of interactions, intervenes slightly less than a teacher. The average number of interactions between the teacher and the student was 10.3 in the Expert Tutoring condition (as shown in Table 2), as opposed to an average of 8.7 interactions in this condition. Peer Tutoring Group 1 Group 2 Group 3 Average
Problems Attempted 1 2 1 1.7
Tutor Interactions 9 7 10 8.7
Student-Solicited Interactions n(%) 5(55.6) 2(28.6) 1(10.0) 2.7(31.4)
Tutor-Initiated Interactions n(%) 4(44.4) 5(71.4) 9(90.0) 6.0(68.6)
Table 3: Tutor mediation for “Peer Tutoring” Summary of Information Flow Findings From studying the flow of information between participants, we have learned that when a teacher or peer tutor is present, information is directed from the tutor to the student. In “Expert Tutoring” and “Peer Tutoring”, the tutor had more knowledge and shared this with the student to direct him or her. In contrast, in “Peer Collaboration” where another same-level peer was available for help, students did not exchange useful information with each other to mediate the learning process; the teacher initiated almost all the interaction in that condition. When two peers are
working together without the presence of a tutor, the peers work independently and do not mediate each other’s learning. In “Collaboration Plus,” when a tutor was present, students still did not collaborate. Instructional Strategies We coded the types of strategies used by tutors, and we found five strategies which are derived from Graesser et.al.’s (1995) categories. Several of his categories were collapsed into a single category for the purpose of this study, since they were general enough to encompass the others. The types of mediation we coded were leading, prompting, hinting, probing, and diagnosing. Diagnosing includes what Graesser and his colleagues identified as “summarizing”, “conducting assessment” and “providing immediate feedback.” Probing includes what they called “pumping” as well as “questioning.” We did not code the “Pure Collaboration” condition because we did not observe information flow between the students in those groups. Figure 4 categorizes the percentage of interventions by both a teacher and a peer-tutor according to the instructional strategies they employed.
50 40 Expert Tutoring
30
Collaboration Plus
20
Peer Tutoring
Diagnosing
Prompting
Leading
0
Probing
10
Hinting
% of Mediation
Tutor Mediation Within Each Condition
Instructional Strategy
Figure 4: Instructional strategies graph Leading Leading differs from prompting in that in a leading strategy, the tutor is specifically directing a student to complete a task. Leading would mean answering a question such that it solves part of the problem, as illustrated in Passage 5. Peer tutors did almost twice as much leading as did teachers. A teacher used similar strategies whether working with one student or two simultaneously. Timmy: I don’t know how to figure that out Tutor: The last column is acres divided by hour. That would be how much work Phillip did in 1 hour.
Passage 5: Example of leading by a teacher Prompting In prompting, tutors (either a peer or teacher) ask questions so that the student can arrive at the answer himself or herself. In Passage 6, the teacher is prompting the student to a particular answer. Prompting is the strategy that the teacher and peer tutor used most often when mediating learning. Tutor: Where would we find the information we need? Timmy: Phillip’s timesheet Tutor: Good, now what does this tell us? Timmy: How many hours Phil worked Tutor: Ok, and what else?
Passage 6: Example of prompting
Hinting Hinting differs from prompting in that hints usually take the form of analogies or reminders and may not relate specifically to the problem. Passage 7 illustrates hinting. The teacher employed the hinting strategy more often that did the peer tutor. Tutor: Remember what rate is? Timmy: Is it the consistency of something? Tutor: Heart rate we defined as the amount of work that your heart does in a given time. Tutor: So Phillip’s rate would be ________?
Passage 7: Example of hinting Probing Probing involves the tutor asking questions to make sure a student understands the problem, as in Passage 8. A peer tutor seems to do less probing than a teacher to make sure an idea or concept is understood. Tutor: You already did his total acres so we just need the total time. Timmy: 19.669 acres in 19.669 total time Tutor: How did you get 19.669 for total time?
Passage 8: Example of probing Diagnosing Diagnosing is a strategy in which tutors provide feedback by evaluating the student’s ideas and actions in the learning environment. As illustrated in Passage 9, diagnosing provides the student with immediate input about their progress. In our study, a teacher did more diagnosing than a peer tutor. Tutor: You are correct – John has mowed 15.75 acres before. Tutor: You are also right that there are 60 minutes in an hour.
Passage 9: Example of diagnosing Summary Primarily, human tutors in a learning environment take the learner through the problem by asking questions pertaining to the problem at hand. The teacher did less probing and diagnosing and more leading and hinting in the “Expert Tutoring” condition than in the “Collaboration Plus” condition. It also appears that a peer tutor prefers prompting followed by leading to mediate student learning. Overall, we found that the notion that a problemsolving partner is more knowledgeable, regardless of his or her status or teaching experience, seems to provide a student with more inclination to ask questions of the partner than in the case where the partner is a same-level student.
DISCUSSION We had hypothesized that students in the “Expert Tutoring” condition would consult the teacher regularly for advice and suggestions. We saw in Table 2 that a student initiated more of the interaction in the case when he or she was able to solve a problem, whereas the teacher initiated more of the interaction when the student was struggling. This would be a logical result of students not knowing when to ask for assistance or what types of questions to ask. In responding to the student, we found that the teacher used prompting most frequently, followed by hinting. It seemed that the teacher expected the students to find the answers on their own with some direction. This is similar to the findings of Lepper et. al. (1993) described earlier that experienced tutors responded to questions and problems indirectly, rather than providing answers. We had hypothesized that students would collaborate as equals in the “Peer Collaboration” condition, working together since they were information resources for each other in the environment. But the students had no prior training in how to collaborate and were not given explicit instructions on how to work together. Perhaps as a result, they proceeded as individual problem-solvers; there was no shared knowledge. Our results support Vedder’s (1985) findings that students working in groups appear to be more focused on finding the right answer than mediating each other’s learning. As in the studies in face-to-face classroom environments (Van der Miej, 1988; Newman et al., 1990), peers did not ask each other for help, possibly due to similar concerns about appearing dependent or
incompetent. Even though the participants in this condition were not complete strangers, they needed support in working together. Next, we had expected in the “Collaboration Plus” condition that we would see the teacher being consulted more often than the other student. However, in most cases, questions or remarks would not be addressed to either the teacher or peer, and the teacher would be responsible for initiating any interaction. Again, this could be a result of students’ perceived costs of help-seeking (Nelson-LeGall, 1992). In “Collaboration Plus,” the instructional strategies used by the teacher were primarily prompting, followed closely by diagnosing. We also noticed that the teacher aimed to facilitate the learning of the student, whereas the peer often aimed just to find the right answer, consistent with Salomon and Perkin’s (1998) work which found that the learner participated more actively when working with an expert tutor instead of a peer. Furthermore, when students did interact, they often misdirected each other or moved off-task, requiring teacher intervention. This is more evidence that the collaboration process needs to be scaffolded in a networked environment. Lastly, we believed that a peer tutor in the “Peer Tutoring” condition would be approached online by a student in a similar manner to a teacher. We found that the peer tutor initiated interaction more often than the student asked for assistance, which is consistent with the study of peer tutoring by McCarthey and McMahon (1992). Because of the nature of the task and the differences in expertise, the tutor assumes more of a traditional teacher role and thus dominates the dialogue. Unlike when the teacher was acting as the tutor, we found that the peer tutor almost half the time chose the more straightforward, simple instructional method of prompting, followed a quarter of the time by leading, another less sophisticated approach, to help a student. Overall, we found that the notion that a problem-solving partner is more knowledgeable, regardless of his or her status or teaching experience, seems to provide a student with more inclination to ask questions of the partner than in the case where the partner is a same-level student. Students benefit from awareness of expertise. This suggests that as long as learners feel confident that collaborators will provide useful suggestions to lead them to a solution, they will willingly ask questions of and answer questions from their partner(s). However, if both a peer and an expert are present, a student is less likely to ask the expert directly for help. This explains why the percentage of tutor-initiated interactions was much higher than student-solicited interactions in the condition where two students and a teacher were present; the teacher often jumped in to offer help because few of the questions were addressed directly to her.
IMPLICATIONS AND CONCLUSION We conclude by returning to our research questions. How does a CSCL learning environment impact the types of support needed to effectively scaffold students? Which elements of peer interaction and teacher interaction are effective in supporting collaborative learning? We cannot make strong claims due to the size of our study, but our results suggest interesting areas for further exploration. Design Implications We noted general design implications from tutor-mediated instruction for designing support for groups of students working in electronic learning environments. In “Expert Tutoring,” we saw that the teacher initiated more of the interaction when the student was struggling. Support for group learning should diagnose student progress and intervene at impasses. This is consistent with the findings of Brown and VanLehn (1988) described earlier. A human or computer tutor does not necessarily need to provide immediate feedback, but a model tracing student progress could ensure that students do not get too far off track. Scaffolds in CSCL environments should also improve communication between students to accelerate their path along an acceptable solution trajectory. We saw in the “Peer Collaboration” condition that students proceeded as individual problem solvers with little useful information shared between them. As suggested earlier in this paper, discussing and explaining ideas to peers is a useful learning strategy. This suggests that supporting collaboration requires •
diagnosing impasses to redirect incorrect solution paths,
•
facilitating problem-solving interaction to keep students engaged, and
•
suggesting ways to divide the problem into sub-tasks for which each participant can be responsible.
These functions are somewhat similar to the findings by Katz and Lesgold (1993) of the role of the tutor in a collaborative situation, except that our results suggest that the tutor in a CSCL environment needs to be more proactive in providing advice. Rather than leaving to students the responsibility for deciding when advice is needed, we believe the tutor needs to identify problems and provide redirection. Finding the optimal balance of peer and expert tutoring strategies is an important research challenge. There are times when a knowledgeable expert is needed, and other times when a peer, to which ideas can be explained and justified, is appropriate. Our study suggests, however, that peers need to be instructed on how to collaborate before they can be effective at facilitating each other’s learning. Experts also need to give students the chance to work together and correct each other before jumping in with the solution. Conclusion More research needs to be done to examine how traditional methods of tutoring differ when the participants are interacting over a computer. Are scaffolding techniques different? Is a human tutor less effective online than faceto-face? More studies should also explore how students learn differently with peers versus teachers in an electronic environment. This study demonstrated that the techniques used to scaffold learning in a CSCL environment are different from interactive learning environments. Not only does individual learning need to be supported, but the interaction between participants needs to be strongly encouraged and facilitated because outside of the social context of the classroom, it is easy for the students to become disengaged. We have performed a small study that has looked more generally at the role of peers, a knowledgeable peer-tutor and a teacher put into an electronic tutoring situation. Our findings suggest that there are significant advantages for both knowledge-rich interactions and peer-to-peer collaboration. Designers of collaborative learning environments need to capture these advantages by developing scaffolds that can effectively lead, prompt, hint, probe, and diagnose in a way that supports team problem-solving without actively leading the team towards a predetermined outcome.
ACKNOWLEDGEMENTS This work was supported by IBM T.J. Watson Research Center. We would like to thank Elizabeth Davis and Nichole Pinkard for their feedback and suggestions. We also thank the participants in this study.
REFERENCES Anderson, J., Boyle, C.F., Corbett, A., & Lewis, M.W. (1990). Cognitive modeling and intelligent tutoring. Artificial Intelligence, 42: 7-49. Anderson, J.R., Corbett, A., Koedinger, K., & Pelletier, R. (1995). Cognitive tutors: Lessons learned. Journal of the Learning Sciences, 4(2): 167-207. Anderson, J.R., & Skwarecki, E. (1986). The automated tutoring of introductory computer programming. Communications of the ACM, 29(9): 842-849. Ayala, G., & Yano, Y. (1998). A collaborative learning environment based on intelligent agents. Expert Systems with Applications, 14: 129-137. Berge, Z. (1997). Computer conferencing and the on-line classroom. International Journal of Educational Telecommunications, 3(1): 3-21. Bloom, B.S. (1984). The 2-sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6): 4-16. Brown, A., & Palincsar, A.S. (1989). Guided, cooperative learning and individual knowledge acquisition. In L. Resnick (Ed.) Knowing, learning, and instruction: Essays in honor of Robert Glaser (pp. 393-451). Hillsdale, NJ: Erlbaum. Brown, J.S., & VanLehn, K. (1988). Repair theory: A generative theory of bugs in procedural skills. In Readings in cognitive science: A perspective from psychology and artificial intelligence. San Mateo, CA: Morgan Kaufmann.
Edelson, D.C., & O'Neill, D.K. (1994). The CoVis collaboratory notebook: Supporting collaborative scientific inquiry. In Recreating the revolution: Proceedings of the National Educational Computing Conference (pp. 146-152). Eugene, OR: International Society for Technology in Education. Graesser, A.C., Person, N.K., & Magliano, J.P. (1995). Collaborative dialog patterns in naturalistic one-on-one tutoring. Applied Cognitive Psychology, 9: 359-387. Gutwin, C., Greenberg, S., & Roseman, M. (1996). Supporting awareness of others in groupware. Proceedings of ACM SIGCHI Conference on Human Factors in Computing Systems. Guzdial, M., Hmelo, C., Hubscher, R., Nagel, K., Newstetter, W., Puntembakar, S., Shabo, A., Turns, J., & Kolodner, J. (1998). Integrating and Guiding Collaboration: Lessons Learned in Computer-Supported Collaborative learning research at Georgia Tech. Retrieved Jan. 4, 2001 from the World Wide Web: http://guzdial.cc.gatech.edu/papers/lessons.pdf. Harasim, L.M., Ed. (1990). Online Education: Perspectives on a New Environment New York: Praeger. Jermann, P., Soller, A., & Muehlenbrock, M. (2001). From mirroring to guiding: a review of state of the art technology for supporting collaborative learning. Proceedings of Proceedings of the First European Conference on Computer-Supported Collaborative Learning,, Maastricht, The Netherlands. Katz, S., & Lesgold, A. (1993). The role of the tutor in computer-based collaborative learning situations. In S.P. Lajoie & S.J. Derry (Eds.), Computers as Cognitive Tools. Hillsdale, NJ: Lawrence Erlbaum Associates. King, A. (1989). Verbal interaction and problem solving within computer aided learning groups. Journal of Educational Computing Research, 5: 1-15. King, A. (1994). Guiding knowledge construction in the classroom: Effects of teaching children how to question and how to explain. American Educational Research Journal, 30: 338-368. Knuth, R.A., & Cunningham, D.J. (1993). Tools for constructivism. In T. Duffy, J. Lowyck & D. Jonassen (Eds.), Designing environments for contructivist learning (pp. 163-187). Berlin: Springer-Verlag. Koschmann, T. (1996). CSCL:Theory and practice of an emerging paradigm. Mahwah, NJ: Lawrence Erlbaum Associates. Lajoie, S., & Lesgold, A. (1989). Apprenticeship training in the workplace: Computer-coached practice environment as a new form of apprenticeship. Machine-mediated learning, 3: 7-28. Lepper, M.R., Woolverton, M., Mumme, D.L., & Gurtner, J.-L., Eds. (1993). Motivational Techniques of Expert Human Tutors: Lessons for the Design of Computer-Based Tutors. Computers as Cognitive Tools. Lesgold, A., Lajoie, S., Bunzo, M., & Eggan, G. (1992). Possibilities for assessment using computer-based apprenticeship environments. In W. Regian & V. Shute (Eds.), Computer assisted instruction and intelligent tutoring systems: Shared issues and complementary approaches. Hillsdale, NJ: Lawrence Erlbaum Associates. McCarthey, S., & McMahon, S. (1992). From Convention to Invention: Three Approaches to Peer Interactions During Writing. In V.K. R. Hertz-Lazarowitz, & N. Miller (Ed.) Interactions in Cooperative Groups. New York: Cambridge University Press. Nelson-LeGall, S. (1992). Children's Instrumental Help-Seeking: Its Role in the Social Acquisition and Construction of Knowledge. In R. Hertz-Lazarowitz & N. Miller (Eds.), Interactions in Cooperative Groups (pp. 49-70). New York: Cambridge University Press. Newman, R., & Goldin, L. (1990). Children's reluctance to seek help with schoolwork. Journal of Educational Psychology(82): 71-80. Nystrand, M. (1986). The structure of written communication: studies of reciprocity between writers and readers. London: Academic Press. Palincsar, A.S., & Brown, A. (1984). Reciprocal teaching of comprehension-fostering and monitoring activities. Cognition and Instruction, 1(2): 117-175.
Rogoff, B. (1990). Apprenticeship in thinking: Cognitive development in social context. New York: Oxford University Press. Salomon, G., & Perkins, D.N. (1998). Individual and Social Aspects of Learning. In P.D. Pearson & A. Iran-Nejad (Eds.), Review of Research in Education. 23. Savery, J.R., & Duffy, T.M. (1996). Problem based learning: An instructional model and its constructivist framework. In B.G. Wilson (Ed.) Constructivist Learning Environments (pp. 135-150). Englewood Cliffs, NJ: Educational Technology Publications. Scardamalia, M., Bereiter, C., McLean, R., Swallow, J., & Woodruff, E. (1989). Computer-supported intentional learning environments. Journal of Educational Computing Research, 5: 51-68. Singley, M.K., Fairweather, P., & Swerling, S. (1999). Team Tutoring Systems: Reifying Roles in Problem Solving. Proceedings of Proceedings of Computer Support for Collaborative Learning, Palo Alto, CA. Teasley, S.D., & Roschelle, J. (1993). Constructing a joint problem space: the computer as a tool for sharing knowledge. In S.P. Lajoie & S.J. Derry (Eds.), Computers as cognitive tools (pp. 229-257). Hillsdale, NJ: Lawrence Erlbaum. Van der Miej, H. (1988). Constraints on question-asking in classrooms. Journal of Educational Psychology, 80: 401-405. Vedder, P. (1985). Cooperative learning. A study on processes and effects of coopration between primary school children. Westerhaven Groningen, The Netherlands: Rijkuniversiteit Groningen. Vygotsky, L. (1978). Mind in society: The development of higher psychological processes. Cambridge, MA: Harvard University Press. Webb, N.M. (1991). Task-related verbal interaction and mathematics learning in small groups. Journal of Research in Mathematics Education, 22: 366-389. Webb, N.M. (1995). Constructive activity and learning in collaborative small groups. Journal of Educational Psychology, 87(3): 406-423.
