Learning from student responses: A domain-independent natural language tutor

Learning from student responses: A domain-independent natural language tutor Jenny McDonald1, Alistair Knott2, Richard Zeng1 and Ayelet Cohen1 1Higher Education Development Centre, University of Otago 2Dept of Computer Science, University of Otago jenny.mcdonald@otago.ac.nz Abstract Providing timely and individualised feedback to students in large undergraduate classes is problematic. In this paper we describe our approach to creating a simple, surface-based, domain-independent natural language tutor which uses simple machine learning tech- niques as a step towards resolving this issue. The focus of our efforts was on developing a high-quality tutorial dialogue plan, creating well-designed questions, and building a model of student responses derived from real student data. We present some early evaluation results and briefly outline the opportunities that our approach and the new tutorial dialogue system present. 1 Introduction The study this paper describes arose out of a practi- cal need expressed by one of the coordinators of an undergraduate first year health sciences course: For large class sizes of 1500-1800 stu- dents, is it possible to use technology to provide timely individualised feedback to students on their understanding of key concepts? Natural language intelligent tutoring systems (ITS) seemed to offer some promise for supporting and en- hancing student understanding of key concepts in this domain. For example, Circsim Tutor (Evens and Michael, 2006) is a natural language tutor de- signed expressly to develop health sciences student understanding of the baroreceptor reflex in humans (the baroreceptor reflex is one of the mechanisms for maintaining blood pressure in humans). Nev- ertheless derision and dismisal of ITS as a failed enterprise are common views among many educa- tional researchers and practising teachers, for exam- ple, Laurillard (2002) and Ramsden (2003). With a few exceptions, even today, within Higher Educa- tion, ITS are hardly in widespread practical use for teaching and learning (Reeves and Hedberg, 2003). There are some good practical reasons for this. Mur- ray (1999), in his review of ITS authoring systems, addresses a key one: Building an explicit model of anything is not an easy task, and requires analy- sis, synthesis, and abstraction skills along with a healthy dose of creativity. ...it is difficult to reduce the entire design task to low level decisions that yield a quality product...some degree of holistic under- standing and abstract thinking will even- tually have to come into play. Worse still, in practice it is seldom feasible to adapt a system designed for a specific teaching and learning context to another. So for example, while Circsim Tutor deals with the baroreceptor reflex it deals with it at a level which is too advanced for the broader introductory-level course on cardiovascular homeostasis that we were dealing with. Even if this were not the case, there would likely be differences in emphasis in terms of the curriculum and adapting a deep system like Circsim would be a non-trivial task. Jenny Mcdonald, Alistair Knott, Richard Zeng and Ayelet Cohen. 2011. Learning from student responses: A domain-independent natural language tutor. In Proceedings of Australasian Language Technology Association Workshop, pages …

Content truncated to 500 words