An Interactive Computer-Based Tutor for LISP

Robert G. Farrell, John R. Anderson, Brian J. ReiserAAAI 1984

Abstract

This paper presents an intelligent tutoring system (ITS) designed to teach LISP programming by emulating the pedagogical strategies of expert human tutors. By integrating a production-system problem-solver with an immediate-feedback advisor, the system significantly enhances learning efficiency compared to traditional classroom settings.

Introduction

Learning a first programming language presents significant cognitive challenges for novices. In the mid-1980s, students typically relied on static textbooks and "cold terminals" that offered little guidance during the problem-solving process. While human tutoring is known to be far more effective than classroom instruction, it is difficult to scale. This research introduces a computer-based tutor that bridges this gap by understanding student intent, providing immediate context-aware feedback, and managing the working memory load inherent in complex programming tasks.

System Architecture and Methodology

The LISP tutor is built upon two core components: the Problem-Solver and the Advisor.

The Problem-Solver utilizes the GRAPES (Goal-Restricted Production System) interpreter. It contains a "novice model" consisting of correct LISP programming rules alongside a "buggy rule catalog"—a collection of common student misconceptions and errors. This allows the system to not only recognize correct code but also diagnose specific logical failures.

The Advisor serves as the interface, monitoring student keystrokes in real-time. It operates across three distinct problem spaces:

  1. Coding Space: A syntax-aware editor for writing LISP code.
  2. Means-Ends Analysis Space: A support mode that helps students bridge gaps between their current state and the desired goal using concrete examples.
  3. Problem Decomposition Space: A conceptual layer that assists students in breaking down complex problems into manageable sub-goals before implementation.

Instructional Design and Interaction

A key innovation of the system is its focus on Interactive Student Modelling. Rather than evaluating only the final output, the tutor monitors the entire development process. By inferring the production rule a student is likely following, the system can intervene the moment a deviation occurs.

To reduce cognitive load, the tutor explicitly displays the current goal hierarchy. This allows students to focus on the immediate logic of a function (e.g., "Test if x is less than y") without losing track of the broader problem structure.

A menu interface used by the LISP tutor to resolve ambiguity in student intent.

When a student's input is ambiguous—meaning it could lead to multiple valid programming paths—the system may prompt the user to clarify their intent. As shown in Figure 1, the tutor provides a menu of possible goals, ensuring the student remains on a productive path while maintaining a clear model of their progress.

Evaluation and Results

Initial experimental evaluations indicated that the LISP tutor was approximately twice as effective as standard classroom instruction. By utilizing the "buggy" rule catalog, the system provided specific, actionable explanations for errors, preventing students from wasting time on unproductive "wandering" through incorrect logic. At the time of publication, 18 hours of instruction had been implemented, with a full 30-hour curriculum planned.

Discussion and Limitations

While highly effective, the system faces challenges with ambiguity and rigidity. If a student's input does not match any known correct or buggy rules, the tutor may struggle to provide guidance. Furthermore, the requirement to stay on a "correct path" may occasionally stifle creative or non-standard solutions that fall outside the system's predefined rule base. Future work focuses on expanding the curriculum and refining cognitive models to better capture the transition from novice to expert programmer.

Summary updated

BibTeX
@inproceedings{farrell1984interactive,
  author = {Robert G. Farrell and John R. Anderson and Brian J. Reiser},
  title = {{An Interactive Computer-Based Tutor for LISP}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1984},
}
Figure