Investigating automatic assessment of reading comprehension in young children

Abstract

This paper describes a preliminary investigation into automatic assessment of reading comprehension in young children. In particular we studied the feasibility of automatic scoring of answers to open-ended questions related to the contents of a passage read by a child. Data from 70 children in grades 1 and 2 were used in this work. An automatic speech recognition system, especially trained for children's speech, was used for tracking the read passage, and two methods for automatic assessment were tested and compared with scores assigned by elementary school teachers. Automatic assessment showed a high kappa statistics agreement with evaluation scores obtained from teachers' scores, K=0.62, comparable to the inter-teacher agreement, K=0.64.