Promoting a Common Testbed for Natural Deduction Tutoring Systems

Abstract

This paper makes the case for common testbeds for evaluating educational systems, such as cognitive tutors, and fully develops a language for specifying such testbeds in a particularly challenging case, that of cognitive tutors for logic and natural deduction. As the teaching of natural deduction spans several disciplines, and notation varies widely, it is impossible for a single specification to apply across all systems, so the key contribution of this paper is defining a "super-language" that we call the gold standard which can capture the capabilities of all currently-used systems, as well as other possible future directions. Translators both to and from individual logic languages allow this to serve as both a specification language for developing standard testbeds, and as an intermediate language to enable simpler system-to-system translation. The development described here can be used as an exemplar for developing common testbeds for other educational systems.