A new method for parsing student text to support computer-assisted assessment of free text answers

Due to current trends in staff-student ratios, the assessment burden on staff will increase unless either students are assessed less, or alternative approaches are used. Much research and effort has been aimed at automated assessment but to date the most reliable method is to use variations of multiple choice questions. However, it is hard and time consuming to design sets of questions that foster deep learning. Although methods for assessing free text answers have been proposed, these are not very reliable because they either involve pattern matching or the analysis of frequencies in a “bag of words”. The first step towards automatic marking of free text answers by comparing the meaning of student answers with a single model answer is to parse the student work. However, because not all students are good at writing grammatically correct English, it is vital that any parsing algorithm can handle ungrammatical text. In this paper, we present preliminary results of using a relatively new linguistic theory, Role and Reference Grammar, to parse student texts and show that ungrammatical sentences can be parsed.