An exploration of the features of graded student essays using domain-independent natural language processing techniques

This paper presents observations that were made about a corpus of 135 graded student essays by analysing them with a computer program that we are designing to provide automated formative feedback on draft essays. In order to provide individualised feedback to help students to improve their essays, the program carries out automatic essay structure recognition and uses domain-independent graph-based ranking techniques to derive extractive summaries. These procedures generate data concerning an essay’s organisational structure and its discourse structure. We have selected 27 attributes from the data and used them in a comparative analysis of all the essays with a view to informing further development of the feedback program. The results of this analysis suggest that some characteristics of students’ essays that our domain-independent feedback program is measuring may be related to the grades that tutors assign to their essays