Refining Border Security News Event Geotagging through Deployment of Lexico-Semantic Patterns

This paper explores a linguistically-oriented method for assigning fine-grained geotagging information to border security-related events reported in online news. It is based on a Cognitive Linguistics model of the semantics of the Locative Prepositional Phrases (LPP) containing place names in news text. We first propose a corpus annotation standard that assigns certain semantic features, such as Locative Relations, to the place names in news articles. Then, based on this, we build a finite state rule-based system that automatically extracts such features and evaluate how they can help discarding implausible event locations. The goal of the experiment is rather a proof of concept that such semantic information can help outperform standard geotagging methods in the border security domain, nonetheless we argument that systems could be built semi-automatically and across languages, based on the proposed annotation standard.