Street-Level Geolocation From Natural Language Descriptions

Nate Blaylock*, James Allen**, William de Beaumont**, Lucian Galescu** and Hyuckchul Jung***
*Nuance Communications; nate.blaylock@nuance.com
**Florida Institute for Human and Machine Cognition (IHMC); allen,wbeaumont,lgalescu,hjung@ihmc.us
***AT&T Labs - Research, Shannon Laboratory; hjung@research.att.com
Résumé (en anglais)
In this article, we describe the TEGUS system for mining geospatial path data from natural language descriptions. TEGUS uses natural language processing and geospatial databases to recover path coordinates from user descriptions of paths at street level. We also describe the PURSUIT Corpus - an annotated corpus of geospatial path descriptions in spoken natural language. PURSUIT includes the spoken path descriptions along with a synchronized GPS track of the path actually taken. Finally, we describe the performance of several variations of TEGUS (based on graph reasoning, particle filtering, and dialog) on PURSUIT.