Marieke van Erp to present NewsReader at Soeterbeeck eHumanities workshop

On Friday June 14, Marieke van Erp will give a talk titled “From events to stories: different ways of structuring the same bag of events over time” at the Soeterbeeck eHumanities Workshop.

The full programme can be found here, and this is the abstract of her talk:

From events to stories: different ways of structuring the same bag of events over time
Marieke van Erp
Postdoctoral researcher at VU University Amsterdam

We know that certain events happened, such as the 2004 Indian Ocean earthquake or the assassination of Pim Fortuyn. However, how these events are described and ultimately end up in the history books, depends on the perspective of the person telling the story. At VU University, three research projects are currently underway that target the extraction of events from text from different perspectives: SemanticOfHistory, BiographyNED and NewsReader. In this talk, I will explain how we extract and model such perspectives and what this means for applications using these perspectives.

In SemanticOfHistory, we generated a model for resolving event coreference that is inspired by the way historical events are described in texts with different historical perspectives. We found that texts closer in time to the event describes these events with more detail and less causal and intentional information. Texts written with more ‘historical’ distance (in time) tend to abstract from details, give a more complete picture and impose more interpretation and opinion. These insights were used to develop a model that matches event descriptions according to the specificity and granularity by which they describe the same event.

In BiographyNED, we are investigating the potential of describing the lives of people in terms of events to discover possible connections but also discover possible evaluations of these events: who are considered to be the bad guys and who are the heroes.

In NewsReader, these models are applied to massive amounts of news streams to record the reporting of events in time. Eventually, this results in a recording of history as told in news media. Such a database not only represents news as a single condensed story showing more systematic connections and relations over time but also shows different versions of these stories as told by different sources.

Comments are closed.