A Blackboard Model for Flexible and Parallel Text Annotation

Creating rich semantic text annotations is a complex process that involves combining multiple natural-language annotation approaches. This annotation process is often approached sequentially and includes pre-processing steps and techniques that build on the outputs of others. However, combining them is not trivial, because some annotation approaches comprise chains of steps or build on other already pre-existing annotations, some pre-processing steps may be common to several techniques, and many newer techniques are even end-to-end which have diluted the need for specific pre-processing steps. Yet it can be beneficial to combine the different approaches because they solve different annotation problems and, even when they solve the same problem, they may have complementary strengths. Whereas existing works often approach the annotation process sequentially, we argue that it can instead be implemented as a partly sequential, partly parallel and concurrent collaboration between independent components. The Blackboard Model is a long-established problem-solving paradigm that deals with complex problems where multiple knowledge sources contribute independently towards the solution. In this work, we study the feasibility of the Blackboard Model for creating rich semantic annotations from text as part of a larger big-data-ready AI system for supporting journalists and newsrooms.