From SDK to xPST: A New Way to Overlay a Tutor on Existing Software

Our past work has investigated the use of the Cognitive Model Software Development Kit (SDK) for creating the cognitive models that underlie model-tracing Cognitive Tutors. Though successful at increasing the number of people who could author such a cognitive model, for certain kinds of situations the Cognitive Model SDK proved cumbersome. The present work discusses a new authoring system, xPST, that allows an example-based tutor to be built on top of existing software.xPST-based tutors have been built for two real-world systems that had existing interfaces. Starting Point: The Cognitive Model SDK We have met with success at developing a SDK for cognitive models (Blessing & Gilbert, 2008). The Cognitive Model SDK allowed authors to develop the representations necessary for the cognitive models for model-tracing intelligent tutors. Carnegie Learning, Inc. uses this SDK to create their Cognitive Tutors for math, a commercially successful intelligent tutoring system. The Cognitive Model SDK enables the creation of cognitive models that contain abstracted instruction over instances. However, we have struggled with its use in other situations. In two of the tutors we had built, for example, the authors created a lot of declarative and procedural representations that ultimately received very little use. For example, Hategekimana, Gilbert, and Blessing (2008) created a tutor for Paint.NET, software similar to Adobe Photoshop. One exercise taught users how to resize and scale an image. While one could imagine using this instruction in multiple image-manipulation instances, in the actual tutor it was used in only a couple of exercises. The power of having a model-tracing tutor, in which the instruction could be abstracted over multiple instances, was lost. However, the author still spent much time creating the representations that contained the instruction. This material is based upon work supported by the National Science Foundation under Grant No. EEC-9876363. Copyright © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. Likewise, when we created the tutor for the CAPE Webbased Authoring Tool used at Vanderbilt University as part of the VaNTH ERC (Roselli et al., 2008), the authoring process contained similar issues. Ultimately, the tutored instruction centered over a set of 8 problems. Much work went into the declarative and procedural structures of the tutor, but their re-use was not nearly as great as what one would see in a Carnegie Learning math tutor. Ultimately, the effort spent developing those representations seemed disproportionate to their usefulness in the completed model. What we desired for these situations was a more streamlined system where the tutoring could be developed without the need for as much underlying structure typical of model-tracing tutors. The result has been the Extensible Problem Specific Tutor (xPST) authoring system. The xPST Authoring System The xPST Authoring System shares some in common with another authoring tool, the Cognitive Tutor Authoring Tool (CTAT; Aleven et al., in press). CTAT allows authors to create Example-based Tutors. These tutors are specific to a certain problem or a narrow class of problems. Likewise, an xPST tutor is specific to a certain problem scenario. The intent of both the CTAT and xPST authoring tools is to allow the author to quickly create a model for a particular problem instance by creating hints and other tutoring aspects while the author manipulates the interface itself. The other goal of xPST is to allow tutoring on any interface. Whereas it is possible to do so with CTAT in some circumstances, typical tutors are built with CTATspecific widgets. The power of xPST’s approach should be apparent. It will enable the creation of a model-tracing-like tutor to be created for any piece of software, opening up possibilities in both academic and commercial settings. The architecture used by xPST to communicate with third-party software is similar to that used by the Cognitive Tutor SDK to communicate with off-the-shelf software. This architecture, called TutorLink, is shown in Figure 1. TutorLink serves as the intermediary between the thirdparty application and the Tutoring Agent (xPST in the current case). It knows how to map actions in the interface 466 Proceedings of the Twenty-Second International FLAIRS Conference (2009)