We are researching automated interpretation of imprecise diagrams, diagrams which contain errors but still convey useful information.
For instance, a sketch map of directions will not be to scale, but it will still help us to find our way. Similarly, a sketch of a software design may contain malformed diagrams, but those diagrams are still useful for identifying some features of the design. Our research is directed towards interpreting such imprecise diagrams and gaining as much information from them as possible.
We are currently using the automated assessment task as our testbed. Assessment is a complex task for which the marker must interpret the intended meaning of a presented answer and award credit accordingly. We have developed systems that mark as well as a human expert, for some simple domains. As the project continues we will extend our work to more complex diagrammatic domains.