Named-entity recognition and data visualization techniques to communicate mission command to autonomous systems

As robotic systems are integrated into mission operations, they provide key benefits over traditional manned systems such as increased endurance, versatility, and risk reduction for personnel. In order for robotic systems to become fully integrated into United States Army missions, they will need to gain a level of autonomy that is closer to that of human personnel. Under Mission Command, the current system of issuing orders, an Operational Order (OPORD) contains the information required to execute a mission. The structure allows for authors of OPORDs to leave many details of the operation open to the discretion of the reader, making task interpretation and execution difficult for an autonomous system. The goal of this project, and one step of many in the process of automating systems to read OPORDs, is to show how OPORDs can be annotated with relevant information (such as locations, coordinates, and organizations) using the natural language processing techniques of tokenization and named-entity recognition (NER). Using A-fold cross validation of a Conditional Random Field (CRF) sequence model on 9 OPORDs containing 38,551 tokens, we were able to extract entities with an overall precision of 0.702, recall of 0.478, and F-measure of 0.569.