Identifying Solution Paths in Cognitive Diagnosis.
Abstract
Consider the problem of describing the thought processes of a human being with respect to how he solves a particular task. How can we infer his knowledge of the task and the procedure or strategy he is using to solve it from his performance? We will call this the problem of cognitive diagnosis. In this paper, we examine the cognitive diagnosis - inferring explanations of observed human behavior. We review previous approaches to cognitive diagnosis, focusing on the methodology of Newell and Simon (1972), which relies on the Newell (1980) problem space hypothesis. This approach proceeds in three stages: (1) identify the problem space in which the subject is operating; (2) identify the subject's path through the problem space; and (3) identify a set of rules that account for the steps along this path. We recount earlier work on automating the diagnostic process, and focus on methods for automating the second of the stages above - identifying the subject's solution path. We describe the Diagnostic Path-Finder ((DPF), an Al system that uses heuristic search to hypothesize a solution path to account for a subject's behavior. Unlike most heuristic search systems, DPF employs psychological criteria to direct its search through the problem space. We examine DPR's performance in explaining errors on multi-column subtraction problems, and compare its explanations to those generated by Brown and Burton's (1978) DEBUGGY system. Finally, we diiscuss the advantages and limitation of our approach to automated cognitive diagnosis. Keywords include: cognitive diagnosis, problem spaces, production systems, machine learning, heuristic search, subtraction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1985
- Accession Number
- ADA152463
Entities
People
- P. Langley
- S. Ohlsson
Organizations
- Carnegie Mellon University