Academic journal article Educational Technology & Society

Meta-Cognitive Strategy Instruction in Intelligent Tutoring Systems: How, When, and Why

Academic journal article Educational Technology & Society

Meta-Cognitive Strategy Instruction in Intelligent Tutoring Systems: How, When, and Why

Article excerpt


Certain learners are less sensitive to learning environments and can always learn; while others are more sensitive to variations in learning environments and may fail to learn (Cronbach & Snow, 1977). We refer to the former as high learners and the latter as low learners. Bloom (1984) argued that human tutors not only raised the mean of test scores, but also decrease the standard deviation of scores. That is, students generally start with a wide distribution in test scores but as they are tutored, the distribution becomes narrower: the students at the low end of the distribution begin to catch up with those at the high end. Another way to measure the same phenomenon is to split students into high and low groups based on their incoming competence then measure the learning gains of both groups. According to Bloom, a good tutor should exhibit an aptitude-treatment interaction: both groups should learn, and yet the learning gains of the low students should be so much greater than those of the high ones that their performance in the post-test ties with that of the high ones. That is, one benefit of tutoring is to narrow or even eliminate the gap between high and low. In order to fully honor the promises of learning environments, an effective system should narrow the gap as much as possible without pulling the high learners down. Many preexisting systems can decrease such differences but not eliminate them. This is due in part to the fact that we do not fully understand why such differences exist.

One of many hypotheses is that low learners lack certain specific skills about how to think, including general problem-solving strategies and meta-cognitive skills. If this hypothesis is true, we expect that teaching students an effective problem-solving strategy would not only improve students' learning gains but also decrease the gap between the low and the high learners. Furthermore, if such problem-solving strategy is domain independent, we expect that learners would learn how to apply the strategy and seek to transfer it to new learning environments. Past research has indicated that these skills can be transferred across domains (Lehman, Lempert, & Nisbett, 1988; Lehman & Nisbett, 1990). However, few studies have investigated transfer of problem-solving strategy across domains.

In this paper, we investigate these questions in a special class of learning environments, intelligent tutoring systems (ITSs) (VanLehn, 2006). We present a study in which two groups of college students studied probability first and then physics. The experimental group studied probability with Pyrenees, an ITS that explicitly taught and required students to employ a general problem-solving strategy (VanLehn et al. 2004); while the control group studied probability with Andes, an ITS that did not teach or require any particular strategy (VanLehn et al. 2005). During subsequent physics instruction, both groups used Andes, which also did not teach or require students to employ any particular strategy.

As reported earlier (Chi & VanLehn, 2007), we found that the experimental group out-performed the control group not only in probability, where the strategy was taught and forced upon the participants, but also in physics where it was not forced upon the participants. Furthermore, the strategy seemed to have lived up to our expectations and transferred from probability to physics. In this paper, we determine whether explicit strategy instruction exhibits an aptitude-treatment interaction, that is, whether it narrows or even eliminates the gap between high and low and, moreover, whether both high and low indeed transfer the strategy to the second domain.


A task domain is deductive if solving a problem requires producing an argument, proof, or derivation consisting of one or more inference steps, and each step is the result of applying a domain principle, operator, or rule. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.