We are using LSA for Assessment of Short Answers in the Domain of Computer Science. LSA is a well established technique for automatic grading of essays and is known to work well on longish texts covering relatively broad domains.
It is not known how well LSA performs on assessing short answers in the context of a more constrained domain. We, as part of the European project, ELeGI, are investigating the extent to which LSA can be a viable, grid-based assessment service for technology enhanced learning. We aim to provide insight into both qualitative and quantitative aspects of the types of questions and short answers that yield acceptable results from an LSA-based automatic marking system.