Automated essay grading: an evaluation of four conceptual models

Automated essay grading has been proposed for over thirty years. Only recently have practical implementations been constructed and tested. This paper describes the theoretical models for four implemented system described in the literature, and evaluates their strengths and weaknesses. All four models make use of comparisons with one or many model answer documents that have been previously assessed by human markers. One hybrid system that makes use of some linguistic features, combined with document characteristics, is shown to be a practical solution at present. Another system that makes use of primarily linguistics features is also shown to be effective. An implementation that ignores linguistic and document features, and operates on the “bag of words” approach, is then discussed. Finally an approach using text categorisation techniques is considered.