This document summarizes research on using natural language processing to extract location information from text for applications in disaster response and earthquake early warning. It discusses the challenges of identifying place names, disambiguating locations, and georeferencing to assign coordinates. Machine learning models have achieved success rates over 90% for detecting spatial relation terms like "near" and "at" and predicting distance ranges. The research is further developing these models with more contextual data. It is part of projects on using sensor networks and citizen reporting to enable early earthquake warnings and inform emergency management.