Identifying Humanitarian Assistance Needs in Low Resource Languages

DARPA's Low Resource Languages for Emergent Incidents  (LORELEI) program [1] is focused on enabling low-cost development of capabilities for low-resource languages, targeted at humanitarian assistance and disaster relief (HADR) in the aftermath of a crisis like an earthquake, tsunami, or epidemic.  In this project we are developing new technologies for quickly ramping up the ability to extract actionable information from online sources related to both population needs (e.g. food or water shortages, lack of shelter, need for medical assistance) and population mental state (e.g. fear, anger).  Methodologically we are focused on advanced topic models and combinations of topic models with deep learning methods.