Analogy Tutor: A Tutoring System for Promoting Conceptual Learning via Comparison

Maria de los Angeles ChangAAAI 2014 · DOI

Abstract

While Intelligent Tutoring Systems (ITS) have proven highly effective for procedural skill acquisition, they often fall short in addressing deep-seated conceptual misconceptions in scientific domains. This paper introduces the Analogy Tutor, an innovative framework designed to bridge the gap between procedural "how-to" knowledge and deep conceptual understanding. Built upon the Companions Cognitive Architecture, the system leverages the Structure-Mapping Theory of analogical reasoning to guide students through instructional comparisons. By identifying structural alignments between scenarios, the tutor facilitates the generation of candidate inferences, helping students discover underlying principles and correct intuitive errors.

Introduction

The success of Intelligent Tutoring Systems in well-defined domains like algebra and computer science is well-documented. However, these systems frequently struggle with conceptual learning, where students may learn to apply formulas without truly understanding the underlying physical or biological principles. For instance, a student might correctly solve a force equation while still maintaining incorrect intuitive beliefs about motion. The Analogy Tutor addresses this by utilizing "instructional comparison"—a powerful cognitive tool for conceptual change. By comparing two related scenarios, students are encouraged to look beyond surface features and focus on the relational structures that define scientific concepts.

The Analogy Tutor Framework

The Analogy Tutor is built on several key AI components that simulate human-like reasoning:

  • Structure-Mapping Engine (SME): At the heart of the system is the SME, which implements Structure-Mapping Theory. It takes structured descriptions of a "base" (a known scenario) and a "target" (a new or problematic scenario) and identifies correspondences between them based on their relational structure (e.g., "If pressure increases, then volume decreases").
  • CogSketch: To facilitate natural interaction, students use CogSketch to draw their scenarios. The system automatically translates these sketches into the symbolic representations required by the SME.
  • Qualitative Reasoning: Instead of relying on complex numerical simulations, the tutor uses qualitative models to represent knowledge. This allows it to reason about causal models in a way that aligns with human mental models, making it more effective for conceptual instruction.

Mechanism of Learning

The central mechanism of the Analogy Tutor is structural alignment. As students compare their sketches, the SME identifies "candidate inferences"—propositions that are true in the base scenario and are likely to be true in the target scenario based on the established mapping.

Figure 1: Visualization of the Structure-Mapping Engine (SME) process, showing correspondences (solid lines) and candidate inferences (dashed lines) between a base and target scenario.

As shown in Figure 1, the system maps nodes and relations from the base to the target. This process allows the tutor to provide highly specific feedback. Rather than simply marking an answer as "wrong," the system can pinpoint exactly where the student's structural mapping has failed, offering hints that guide the student to "fill in the blanks" of their own understanding. This domain-independent reasoning suggests that a single AI architecture could potentially teach diverse subjects, from thermodynamics to evolutionary biology.

Evaluation and Future Directions

Initial development has focused on creating qualitative causal models for physics. The proposed evaluation plan involves testing the system's ability to interpret student-generated sketches across multiple domains and conducting controlled studies to measure learning gains compared to traditional instructional methods.

Future work will focus on making the hint-generation mechanism more domain-independent and improving the robustness of the sketch-to-logic translation. By refining the structure-mapping task model, the Analogy Tutor aims to provide even more granular feedback, potentially serving as a general-purpose architecture for teaching complex conceptual knowledge across diverse scientific fields.

Conclusion

The Analogy Tutor represents a shift from procedural-focused AI tutoring toward systems capable of fostering deep conceptual change. By leveraging the domain-independent power of analogical reasoning, this framework offers a promising path toward more flexible and effective intelligent tutoring systems.

Summary updated

BibTeX
@inproceedings{chang2014analogy,
  author = {Maria de los Angeles Chang},
  title = {{Analogy Tutor: A Tutoring System for Promoting Conceptual Learning via Comparison}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2014},
  doi = {10.1609/aaai.v28i1.8784},
}
Figure