Inferring Spatio-Temporal Trajectories of Entities from Natural Language Documents

点击翻译按钮获取中文摘要

This exploratory research focuses on the development of algorithms for integrating spatial and temporal annotations over natural language text, thereby enabling the tracking of entities through space and time. This involves the use of lexical resources and the integration of two existing annotation schemes to create a representation capturing the movement of individuals through spatial and temporal locations. This representation is extracted automatically from documents using symbolic and machine learning methods.This work builds on technologies that have emerged recently that parse the temporal structure of narratives. These techniques use the TimeML markup language to combine rule-based systems, machine learning, and temporal reasoning, and a markup scheme called SpatialML to map relative and absolute locations to geo-coordinates. Data structures from these schemes are then integrated with a representation of event arguments. Using a verb lexicon that captures the meaning of motion verbs, information about the participants involved in events described by such verbs is captured. Finally, these markup representations are mapped onto the appropriate ontological categories within the Standardized Upper Model Ontology (SUMO). The results of this exploratory research are potentially significant, as there has to date been little research done on integrating spatial information extraction with other aspects of text understanding. Furthermore, by providing a mapping of the representations of temporal and spatial annotations over natural language texts to a standardized ontology such as SUMO, we hope to provide interoperability of resources to the community, while also leveraging the work done within the ontology research community.