Refinement

Theory Committee Members

Learners must often revise their knowledge-they add, subtract or modify knowledge components and associated features based on evidence from their instruction. Sometimes the instruction is so direct that it hardly qualifies as "evidence," as when an instructor provides explicit verbal descriptions like telling a student that the French word "jeux" is plural. Sometimes students gather evidence themselves in examples, which may be "unlabeled", such as observing that "jeux" always appears with plural articles and verbs. Students can also get feedback during practice, such as having "le jeux" corrected to "les jeux" in one's essay. Direct instruction and feedback provide scaffolding that aids an inductive knowledge refinement process that would otherwise be quite lengthy. Although they save the learner time by directing much of the refinement process (e.g., discriminating deep from shallow features) for the learner, they risk making the learner dependent upon the instruction for future knowledge refinement. Students may thus not develop meta-cognitive processes for self-supervised learning.

Although many wars in education have been fought over how direct instruction should be, we focus on understanding the tradeoffs between short-term and long-term gains that arise with different methods for scaffolding knowledge refinement. This venerable challenge can be met with the aid of our LearnLabs, which allow observing both short-term and long-term effects.

A fundamental issue of knowledge refinement is exploring the continuum between having the learner do most of the knowledge refinement process and having the instruction do most of the knowledge refinement process. Anything that requires significant work from human students is likely to turn out differently in the classroom than in the laboratory because the motivations and contexts are so different. Thus, it is essential that practitioners and experienced instructors in particular, collaborate with researchers who have done most of their work in the lab. We expect bidirectional payoffs as the laboratory scientists inspire novel practical methods and the domain instructors inspire theoretically interesting methods.

This cluster is strongly related to the co-training cluster. They share an interest in exploring multiple methods drawing from both machine learning and human learning. The co-training cluster focuses on combining them, whereas the knowledge refinement cluster focuses on comparing them. Both are interested in theoretical implications, and in fleshing the implications out as simulated students.

A number of studies in other clusters address issues of knowledge refinement and particularly how best to achieve feature validity. For instance, Ogan et al.'s Cultural Learning study (discussed in the Dialogue Cluster) addresses the refinement cluster by testing whether we can assist students to "add, subtract or modify their beliefs" about a subject in which little explicit instruction is typically given and many unfounded assumptions or expectations may be confronted. It adapts a feature validity technique, guided noticing of important features (in video clips), from a classroom activity to an interactive on-line tutor.