At NewsReader we’ve been working hard since the kick-off to set up the infrastructure for the project. One of the most important things that we needed to get done as soon as possible was to agree on an annotation format, so we can get going on building an evaluation data set.
The result of close collaboration between the research partners resulted in the Grounded Annotation Framework (GAF). This framework consists of two layers, linguistic layer, and a semantic layer, to connect linguistic annotations to semantic denotations. You may wonder, yet another framework, but all annotation frameworks we investigated were either purely linguistic, or purely Semantic Web, and none satisfied our requirements for being flexible, as well as powerful. However, GAF is based on prior frameworks for event annotation, and utilises accepted semantic web standards to represent the semantic layer. Combined with ongoing efforts in the NewsReader project on interoperability, we expect GAF to also be compatible with other frameworks. We will present the GAF framework at the 1st Workshop on EVENTS: Definition, Detection, Coreference, and Representation. This workshop takes places in conjunction with NAACL-HLT in Atlanta, GA, USA on 14 June.
For more information, read the preprint of our paper, or skip the GAF website if you prefer a more practical introduction.