Automatic generation of fine-grained representations of learner response semantics

Abstract

This paper presents a process for automatically extracting a fine-grained semantic representation of a learner's response to a tutor's question. The representation can be extracted using available natural language processing technologies and it allows a detailed assessment of the learner's understanding and consequently will support the evaluation of tutoring pedagogy that is dependent on such a fine-grained assessment. We describe a system to assess student answers at this fine-grained level that utilizes features extracted from the automatically generated representations. The system classifies answers to indicate the student's apparent understanding of each of the low-level facets of a known reference answer. It achieves an accuracy on these fine-grained decisions of 76% on within-domain assessment and 69% out of domain. © 2008 Springer-Verlag Berlin Heidelberg.