Albert T. Corbett

Cognitive modeling and intelligent tutoring

Abstract : The ACT theory of skill acquisition and its PUPS successor provide production-system models of the acquisition of skills such as LISP programming, geometry theorm-proving, and solving of algebraic equations. Knowledge begins in declarative form and is used by analogical processes to solve specific problems. Domain specific productions are compiled from the traces of these problem solutions. The model-tracing methodology has been developed as a means of displaying this cognitive theory in intelligent tutoring. Implementation of the model-tracing methodology involves developing a student model, a pedagogical module, and an interface. issues associated with the development of each of these components are discussed. Work on tutoring and work on skill acquisition have proven to symbiotic; that is, each has furthered the others development. Keywords: Analogy; Artificial Intelligence; Cognitive Science; Computer-Assisted Instruction.

Skill acquisition and the LISP tutor

An analysis of student learning with the LISP tutor indicates that while LISP is complex, learning it is simple. The key to factoring out the complexity of LISP is to monitor the learning of the 500 productions in the LISP tutor which describe the programming skill. The learning of these productions follows the power-law learning curve typical of skill acquisition. There is transfer from other programming experience to the extent that this programming experience involves the same productions. Subjects appear to differ only on the general dimensions of how well they acquire the productions and how well they retain the productions. Instructional manipulations such as remediation, content of feedback, and timing of feedback are effective to the extent they give students more practice programming, and explain to students why correct solutions work.

The knowledge-learning-instruction framework: bridging the science-practice chasm to enhance robust student learning.

Despite the accumulation of substantial cognitive science research relevant to education, there remains confusion and controversy in the application of research to educational practice. In support of a more systematic approach, we describe the Knowledge-Learning-Instruction (KLI) framework. KLI promotes the emergence of instructional principles of high potential for generality, while explicitly identifying constraints of and opportunities for detailed analysis of the knowledge students may acquire in courses. Drawing on research across domains of science, math, and language learning, we illustrate the analyses of knowledge, learning, and instructional events that the KLI framework affords. We present a set of three coordinated taxonomies of knowledge, learning, and instruction. For example, we identify three broad classes of learning events (LEs): (a) memory and fluency processes, (b) induction and refinement processes, and (c) understanding and sense-making processes, and we show how these can lead to different knowledge changes and constraints on optimal instructional choices.

Locus of feedback control in computer-based tutoring: impact on learning rate, achievement and attitudes

The advent of second-generation intelligent computer tutors raises an important instructional design question: when should tutorial advice be presented in problem solving? This paper examines four feedback conditions in the ACT Programming Tutor. Three versions offer the student different levels of control over error feedback and correction: (a) immediate feedback and immediate error correction; (b) immediate error flagging and student control of error correction; (c) feedback on demand and student control of error correction. A fourth, No-tutor condition offers no stepby-step problem solving support. The immediate feedback group with greatest tutor control of problem solving yielded the most efficient learning. These students completed the tutor problems fastest, and the three tutor-supported groups performed equivalently on tests. Questionnaires revealed little student preference among the four conditions. These results suggest that students will need explicit guidance to benefit from learning opportunities that arise when they have greater control over tutorial assistance.

Adapting to when students game an intelligent tutoring system

It has been found in recent years that many students who use intelligent tutoring systems game the system, attempting to succeed in the educational environment by exploiting properties of the system rather than by learning the material and trying to use that knowledge to answer correctly. In this paper, we introduce a system which gives a gaming student supplementary exercises focused on exactly the material the student bypassed by gaming, and which also expresses negative emotion to gaming students through an animated agent. Students using this system engage in less gaming, and students who receive many supplemental exercises have considerably better learning than is associated with gaming in the control condition or prior studies.

Cognitive Computer Tutors: Solving the Two-Sigma Problem

Individual human tutoring is the most effective and most expensive form of instruction. Students working with individual human tutors reach achievement levels as much as two standard deviations higher than students in conventional instruction (that is, 50% of tutored students score higher than 98% of the comparison group). Two early 20th-century innovations attempted to offer benefits of individualized instruction on a broader basis: (1) mechanized individualized feedback (via teaching machines and computers) and (2) mastery learning (individualized pacing of instruction). On average each of these innovations yields about a half standard deviation achievement effect. More recently, cognitive computer tutors have implemented these innovations in the context of a cognitive model of problem solving. This paper examines the achievement effect size of these two types of student-adapted instruction in a cognitive programming tutor. Results suggest that cognitive tutors have closed the gap with and arguably surpass human tutors.

Developing a generalizable detector of when students game the system

Some students, when working in interactive learning environments, attempt to “game the system”, attempting to succeed in the environment by exploiting properties of the system rather than by learning the material and trying to use that knowledge to answer correctly. In this paper, we present a system that can accurately detect whether a student is gaming the system, within a Cognitive Tutor mathematics curricula. Our detector also distinguishes between two distinct types of gaming which are associated with different learning outcomes. We explore this detector’s generalizability, and find that it transfers successfully to both new students and new tutor lessons.

Simplifying video editing using metadata

Digital video is becoming increasingly ubiquitous. However, editing video remains difficult for several reasons: it is a time-based medium, it has dual tracks of audio and video, and current tools force users to work at the smallest level of detail. Based on interviews with professional video editors, we developed a video editor, called Silver, that uses metadata to make digital video editing more accessible to novices. To help users visualize video, Silver provides multiple views with different semantic content and at different levels of abstraction, including storyboard, editable transcript, and timeline views. Silver offers smart editing operations that help users resolve the inconsistencies that arise because of the different boundaries in audio and video. We conducted a preliminary user study to investigate the effectiveness of the Silver smart editing. Participants successfully edited video after only a short tutorial, both with and without smart editing assistance. Our research suggests several ways in which video editing tools could use metadata to assist users in the reuse and composition of video.

Pronoun disambiguation: Accessing potential antecedents

Two general classes of pronoun disambiguation processes are considered. In reading “Jack threw a snowball at Phil, but he missed,” both possible antecedents of “he” (“Jack” and “Phil”) may be accessed initially. Or, the actual antecedent alone may be accessed after sufficient semantic context is encoded. To evaluate these alternatives, a yes-no-probe recognition task was used to measure priming of the potential antecedents in sentence comprehension. Subjects read sentences similar to the example and were presented a test word immediately following each sentence. Response times for the actual antecedent (“Jack”) and nonantecedent (“Phil”) probes were obtained. Results indicated that the nonantecedent as well as the antecedent was activated (accessed) in pronoun disambiguation. This conclusion was not affected by the ordering of the antecedent and nonantecedent in the first clause.

Instrument inferences in sentence encoding

Reading comprehension is an active inferential process. What factors control inferences? This paper examines the possibility that highly probable inferences are drawn even when they are unnecessary for comprehension. Three experiments are reported in which subjects read a list of sentences, each describing an action that accepts a highly likely instrument (e.g., “shovel” for “digging a hole”). In some sentences an explicit instrument is presented. In others the instrument remains implicit. Instrument cue effectiveness is then examined in a subsequent cued recall task. It is concluded that recall cue effectiveness is not a reliable measure of encoding in comprehension. It is further concluded, on the basis of instrument recall, that highly probable implicit instruments are not routinely inferred in comprehension.