Streamlining the Production and Visualization of Error-Free Geospatially Grounded Text Documents

We propose to streamline the production of `self-mapping' documents through the tight integration of evidence-driven automated tagging with a mixed-initiative approach to tag validation and editing. The proposed system, GeoDoc, will deviate substantially from the traditional corpus driven approaches to geospatial metadata generation in order to address the unique issues involved in producing error-free geospatially grounded documents. In particular, we will address accuracy issues by adopting a new model of evidence integration that will allow us to move away from the current practice of ad hoc heuristic combination while exploiting extra-linguistic indicators when available. With that said, the design of GeoDoc is predicated on the fact that within this task, the ultimate measure of system efficacy is the level of human-effort required to produce error-free documents. In particular, we will exploit our reasoning process to support a new cooperative model of editing wherein the system orders editing decisions so as to reduce overall uncertainty in its tagging and thereby reduce document validation time. Documents produced with GeoDoc will support new modes of visualization(and other services), offering substantial value to the information consumer. Phase I will prove the feasibility of our overall approach via prototype.