Enhancing Example-Based Learning in Hypertext Environments

In previous research, Gerjets, Scheiter, and Tack (2000) demonstrated that learners experience serious difficulties in utilizing instructional examples according to their profitability when interacting with a hypertext-based learning environment. In this paper we focus on possible causes of these difficulties and on different instructional methods for improving learners' utilization of worked-out examples in hypertext environments. The results of two experimental studies are reported. Learning from Worked-Out Examples: The Role of Example Processing Strategies and Example Design Research over the last 15 years in the domain of learning and problem solving has demonstrated that instructional examples play an important role for knowledge acquisition in domains like mathematics, physics, or programming (Chi, Bassok, Lewis, Reimann, & Glaser, 1989). In particular for initial skill acquisition, learning from worked-out examples seems to be superior to actively solving training problems (Sweller, van Merriënboer, & Paas, 1998). However, numerous findings also indicate major drawbacks of example-based learning. In particular, poor learners tend to overuse examples during problem solving without reflecting on their appropriateness (VanLehn & Jones, 1993). In addition, learners have difficulties identifying relevant information in worked-out examples and are often distracted by examples' surface features (Ross, 1989). Furthermore, Renkl (1999) assumes that students often suffer from illusions of understanding when learning from worked-out examples. I.e., they may have the false impression of having grasped the solution rationale of an example problem. Finally, learners have difficulties generalizing solutions from examples to novel problems (Catrambone & Holyoak, 1989; Reed, Dempster, & Ettinger, 1985). A number of empirical studies have identified features of example processing strategies and example design that are efficient for successful knowledge acquisition (cf. Atkinson, Derry, Renkl, & Wortham, 2000). • Important strategical aspects mainly concern the adequate selection and elaboration of instructional examples. Reed, Ackinclose, and Voss (1990) showed that learners failed to select sufficiently complex instructional examples for learning although the profitability of these examples for subsequently solving test problems could be demonstrated. However, Reed, Willis, and Guarino (1994) found that learners who were allowed to select worked-out examples while solving test problems were able to select suitable examples. Additionally, it has been shown that selfexplanations are an important aspect of good learners’ example processing (Chi et al., 1989; Pirolli & Recker, 1994; Renkl, 1997). In particular, anticipations of solution steps and inferences with regard to the relations between solution steps, goals, and abstract principles have been proven useful for knowledge acquisition. • With respect to design issues it could be shown that multiple examples can support schema induction which helps learners to solve novel problems (Cummins, 1992). Providing multiple examples with different surface features might further improve this process of abstraction (Quilici & Mayer, 1996). Additionally, it has been proposed that the provision of completion problems where learners have to fill in some details of worked-out examples’ solution steps is a helpful instructional device as it fosters self-explanations (Van Merrienboer, 1990). In particular, presenting completion problems along with evaluative feedback on subjects’ gap-filling performance seems to improve learning outcomes. For instance, Stark (1999) showed that learners benefit from such a combination of completion problems and feedback and stresses the point that completion problems foster example elaboration whereas giving feedback on the learning success might prevent learners from illusions of understanding. From these findings on learning from examples it can be argued that strategies of example selection and processing as well as features of example design have to be taken into account to improve learning outcomes. The aspect of adopting suitable strategies gains increasing importance the more the control of the learning process is left up to the learner. In learning situations where the learner can select instructional material as well as determine the sequence and the pace of presentation, the importance of strategies increases (Gerjets, Scheiter, & Tack, 2000). Therefore, an identification of suitable strategies of information utilization and an examination of whether learners can adopt these strategies is highly relevant when more focus is put on self-regulated learning in the field of instruction. Example-Based Hypertext Environments One domain in which these issues of learner control are stressed is the field of hypertext-based learning where the user can select among different kinds of information and where he can choose according to his goals when the information is to be presented and in which order (Rouet & Levonen, 1996). On the one hand, this allows for great flexibility and adaptivity of learning and problem solving. Generally, it is assumed that the nonlinear structure of hypertext environments improves learners’ ability to use knowledge in a flexible way, so that they learn to apply one information unit to serve different purposes in a variety of situations. Non-linearity also enables learners to utilize information units according to their goals and to their prior knowledge. With regard to example-based learning, providing multiple examples with different surface features in a nonlinear hypertext environment allows the learner to compare examples within one problem category as well as to compare examples between different problem categories. These comparisons are fundamental for processes of abstraction in that they allow learners to identify structural features that define different problem categories. Therefore, non-linearity and the resulting opportunities of flexible information utilization may be especially suitable when learning from examples. On the other hand, learners can "face new problems in selecting and accessing relevant information" (Rouet, Levonen, Dillon, & Spiro, 1996, p. 3). Problems can arise if learners do not possess the necessary prerequisites to cope with the demands that have been imposed to them by redirecting control over the learning process to them (Rouet & Levonen, 1996). Learning with a nonlinear hypertext increases the amount of control demands by making it necessary that learners permanently make decisions about the profitability of individual information units with regard to their current learning tasks. Even if all information provided is relevant to the current task, the information items may differ with respect to their profitability in terms of their processing costs and their contribution to improving the learning outcome (cf. Pirolli & Card, 1999). Therefore, learners may have to develop adequate strategies of information selection and processing in order to make use of the potential benefits of hypertext-based information presentation. Based on these considerations a question of central importance in example-based learning with hypertext is whether learners are capable of utilizing examples according to their profitability, i.e., select, sequence, and compare them in a suitable way. Most research on learning from examples up to now has focused on learning situations where learners have been forced to process the examples provided in a predefined sequence and, in some studies, even for a fixed amount of learning time. However, it is not clear whether these findings can be easily transferred to more natural learning situations that allow subjects to select information in different sequences and to control their own pace of studying. Results of Previous Experiments In a series of previous experiments Gerjets, Scheiter, and Tack (2000) demonstrated that learners experience difficulties in hypertext environments with regard to their ability to utilize examples according to their profitability. These experiments were conducted using a web-based hypertext environment for training and testing in the domain of combinatorics (HYPERCOMB). During the learning phase subjects could retrieve abstract information on six problem categories from the domain of combinatorics. Depending on the experimental condition, this abstract information was either not augmented by any additional instructional information or was augmented by one or three worked-out examples that illustrated the six problem categories. Learners could retrieve the information they wanted to study and could determine the pace and sequence of information presentation. When they had the impression that they had learned sufficiently well learners could switch to a test phase where they had to solve three test problems. Automated logfile analyses were used to track subjects’ strategic navigation behavior. Additionally, subjects’ problem-solving performance was registered. In order to investigate strategic adaptation to different instructional situation Gerjets et al. (2000) studied learners with either low or high domain-specific prior knowledge using different instructional versions of HYPERCOMB (no or one example or three examples per problem category) with or without time pressure. As a result of their experiments Gerjets et al. (2000) showed that learners have difficulties in selecting the most profitable information in a specific instructional situation. A comparison among the three instructional conditions yielded no beneficial effects of merely providing examples compared to providing only abstract information. However, if subjects made use of the examples in a suitable way (e.g. by comparing different examples) this clearly improved their learning and problem solving performance compared to subjects who made insufficient use of the instructional material. These findings on information profitability were contrasted with learners' actual information utilization behavior. Despite the fact that example processing proved to be useful, about half of the subjects demonstrated poor example processing strategies as they neither processed each example in the one-example condition more than once nor did they study more than one example per problem category in the three-example condition. Hypotheses: Possible Explanations for Learners’ Problems to utilize instructional examples according to their profitability There might be two different explanations for learners’ problems to utilize instructional examples according to their profitability, which will be described in the following paragraphs. Furthermore, two experimental conditions will be outlined that were designed to counteract these hypothesized causes of subjects’ failures in using examples adequately.