In this paper we introduce a taxonomy of classroom discourse with particular focus on mathematical problem-solving discourse. We first discuss the hierarchical nature of classroom discourse and describe how our taxonomy addresses this hierarchical structure. We then describe an approach to classroom discourse classification based on our proposed taxonomy using Conditional Random Fields with features originating from multiple linguistic levels. The multilevel features reduce the classification error rate by over 40% compared with a purely unigram lexical features baseline. The framework and approach proposed in this paper can be useful in future work in education research, as well as discourse analysis research and intelligent tutoring applications.
By: Juan M. Huerta
Published in: RC24870 in 2009
Questions about this service can be mailed to reports@us.ibm.com .
