A linguistically-driven methodology for detecting impending and unfolding emergencies from social media messages

Natural disasters have demonstrated the crucial role of social media before, during and after emergencies (Haddow & Haddow 2013). Within our EU project Sland ´ ail, we aim to ethically improve ´ the use of social media in enhancing the response of disaster-related agen-cies. To this end, we have collected corpora of social and formal media to study newsroom communication of emergency management organisations in English and Italian. Currently, emergency management agencies in English-speaking countries use social media in different measure and different degrees, whereas Italian National Protezione Civile only uses Twitter at the moment. Our method is developed with a view to identifying communicative strategies and detecting sentiment in order to distinguish warnings from actual disasters and major from minor disasters. Our linguistic analysis uses humans to classify alert/warning messages or emer-gency response and mitigation ones based on the terminology used and the sentiment expressed. Results of linguistic analysis are then used to train an application by tagging messages and detecting disaster- and/or emergency-related terminology and emotive language to simulate human rating and forward information to an emergency management system.