The Astus Framework

Intelligent Tutoring Systems (ITS) have been proven successful, but their success is mitigated by the amount of effort needed to build them. Typically, the main factor behind these efforts is the model of the task. Different approaches have been investigated to reduce these efforts: frameworks for Model-Tracing Tutors (e.g. Cognitive Tutors) or Constraint-Based Tutors (e.g. ASPIRE), Example-Tracing Tutors (e.g. CTAT, ASSISTment) and machine learning-based tutors (e.g. SimStudent).

 With Astus, we aim to offer a framework to the ITS community to support the creation of Model-Tracing Tutors for a wide range of well-defined domains. In such context, building a framework based on a generative model of the task was deemed the most interesting approach because it appeared as the only one leading to comprehensive, flexible and reusable pedagogical behaviors. For instance, the tutor is able to give next-step hints that are generated using domain-independent templates instead of domain-specific canned ones. Such flexibility is especially useful to conduct experiments that compare the effects of different pedagogical strategies on the learners.

 Astus’s knowledge representation system is based on interpretable knowledge elements that represent tutored skills and black-box knowledge element that operationalize the underlying mastered skills. Using an authoring language, the model is encoded in a format very similarly to typical source files. Tools for debugging and visualization are available at runtime. 

As the trending approaches are to eliminate the need for a model of the task, or to ask for non-programmers to create one, but instead we are investigating an approach that consists in reducing the required efforts by the programmers. We are aware that our approach, based on a generative model of the task, may be justified only in complex, but well-defined tasks, but doing so, we hope that we can shed a new light on Model-Tracing tutors.

 Our first step was to reproduce tutors built with a comparable framework. For instance, we replicated a TDK (Cognitive Tutor) scatter plot tutor and we simultaneously created a multi-column subtraction tutor using CTAT (a Jess-based Cognitive Tutor) and Astus. Our second step was to conduct experimentations with undergraduates from our department. We made a floating-point number conversion tutor for a system programming course and an AVL-tree tutor for a data structures course. Other projects follow below.

Critical care

As part of the GRITS, we collaborated with the nursing science school (Jetté, S. and Talbot, T.) to create a tutor for a critical care course (TNCC).

Genetic engineering

We collaborated with the microbiology professor Claude Déry to create a tutor that simulates the experimental restriction map procedure.

The technologies Astus depends on :