Conference,
Data-driven Modelling of Human Tutoring in Calculus
(2006)
Abstract
We describe results from analyzing a corpus of human-human tutorial dialogue, which are aimed at identifying, formalizing and automating natural tutoring in the domain of calculus, in particular symbolic differentiation. We analyzed a corpus of human-human tutoring dialogues, focusing on three areas important to system development: interleaved symbolic and natural language, domain modeling, and tutorial dialogue strategies. We provide empirical verification of previous results on interleaving natural and symbolic language, and show that the properties of interleaving are highly dependent on the input modality. We describe a task model for our domain, and provide corpus data to show that the model must cover basic algebra skills as well, which are involved in differentiation. We verify the applicability of an existing annotation scheme for tutoring algebra in our domain, and propose that it be extended to cover student initiative.
