Automatic Extraction of Archaeological Events from Text

This paper describes a series of experiments to automatically detect and categorise archaeological events—such as survey, excavation, finds and so forth—that are described in natural language text documents. Complex event structures with attributes including date, agent and location are extracted and converted into families of binary relations. These in turn can be mapped to RDF triples for publication as Semantic Web graphs, with the potential of making it dramatically easier to interconnect separate data silos. We present results indicating that although events do not conform to the standard definitions of “entities”, they can be detected with high precision, making large-scale processing of text documents a practical possibility.