Automated Journalism 2.0: Event-driven narratives
David Caswell, Konstantin Nicholas Dörr · Journalism Practice · 2017
This article introduces an exploratory computational approach to extending the realm of automated journalism from simple descriptions to richer and more complex event-driven narratives, based on original applied research in structured journalism. The practice of automated journalism is reviewed and a major constraint on the potential to automate journalistic writing is identified, namely the absence of data models sufficient to encode the journalistic knowledge necessary for automatically writing event-driven narratives. A detailed proposal addressing this constraint is presented, based on the representation of journalistic knowledge as structured event and structured narrative data. We describe a prototyped database of structured events and narratives, and introduce two methods of using event and narrative data from the prototyped database to provide journalistic knowledge to a commercial automated writing platform. Detailed examples of the use of each method are provided, including a successful application of the approach to stories about car chases, from initial data reporting through to automatically generated text. A framework for evaluating automatically generated event-driven narratives is proposed, several technical and editorial challenges to applying the approach in practice are discussed, and several high-level conclusions about the importance of data structures in automated journalism workflows are provided.