A human-computer collaborative approach to the marking of free text answers

We propose the term Human-Computer Collaborative Assessment (HCCA) for a distinct and currently rather neglected sub-field of CAA. In HCCA answers are constructions rather than selections, and marking is a process of active collaboration between human marker and machine. We present the results of experience with simple tools which demonstrate significant time savings compared to traditional paper marking. Further improvements in both speed and quality of assessment are clearly possible, but require much more sophisticated tools, particularly for free text answers. We review the role which Natural Language Processing techniques can play, particularly in the light of experience from other domains. Analysis of a number of answer sets highlights key issues in HCCA as well as underlining the infeasibility of fully automatic marking in many situations.