Linking Cognitive Science to Education: Generation and Interleaving Effects

Performance during a learning event is frequently used as a measure of learning; however, basic cognitive research suggests that this may be an unreliable predictor of long-term learning and transfer. Rather, in some training paradigms, higher error rates during training may predict greater retention and generalization of learning. One such paradigm is training in which learners generate, rather than read, during study opportunities. A second is training in which study for two sets of information is interleaved rather than grouped into separate blocks. Educationally relevant learning requires retention and the capacity to generalize information across contexts, leading to the hypothesis that these paradigms may have important implications for educational practice. An experiment is described in which the effects of generation and interleaving are examined using complex, educationally-relevant materials. Findings indicate that these effects are relevant for instructional design, but that there is not a straightforward relationship between laboratory research with simple materials and educational practice. Rather, the educational goal must be considered when determining the utility of generation and interleaving principles in designing instructional technology. A large body of research on cognition exists in which simple materials have been used to understand the mechanisms underlying learning, retention, and transfer. Such research has led to rich understanding of principles that guide learning in laboratory contexts, but less is known about how these principles generalize to learning of the more complex content typically acquired in educational contexts. This paper describes a study that links laboratory research on learning with educational practice. In particular, this study explores “desirable difficulty” (Bjork, 1994, 1999) findings that have been well studied in laboratory contexts and that may have important, albeit unintuitive, implications for educationally-relevant learning. A common strategy for optimizing student learning in classroom contexts is the manipulation of study conditions to increase students’ ease of comprehension and to improve students’ performance on assessments embedded into the learning task. This is a strategy based on the assumption that performance during learning is a reliable index of knowledge and will predict learning and memory for the material over time. This assumption is made regularly by instructors as well as by learners themselves. Children and adults examine their own performance as a metacognitive strategy when making predictions about their own learning and retention over time (Benjamin & Bjork, 1996; Koriat & Bjork, in press) and when determining how to structure study time or when making decisions about how to optimize learning events (Son, under review). While the relationship between performance and learning is regularly invoked in both educational practice and research on learning, other cognitive research indicates that this assumption may be misleading. A growing body of laboratory research indicates that increasing the difficulty of a learning activity in systematic ways may impair learners’ performance during initial knowledge acquisition but improve learning and retention for this learning over time. Several specific principles have been identified and defined as “desirable difficulties” (Bjork, 1994, 1999). Desirable difficulties are specific types of manipulations of the learning context that increase learners’ errors during learning and slow initial acquisition of materials but lead to increased memory for the relevant information over time. Generation Generation is an instructional strategy that impairs initial knowledge acquisition and slows learning but leads to increased likelihood of retention for learned information. The principle of generation describes the reliable phenomenon that having learners generate components of a learned stimulus produces better learning and retention that having learners read or re-read study materials. For example, Hirshman & Bjork (1988) found that requiring a learner to generate letters within the second word in a pair such as