Cross-Lingual Unsupervised Extraction (CLUE)

The goal of this Direct to Phase II effort is to develop a prototype system for Cross-Lingual Unsupervised Extraction (CLUE) that can extract relevant information from across a variety of sources, modalities, and languages.  The CLUE system will extract entities and relations of interest from the different sources and languages, linking this information across documents, associating it with relevant geospatial information, and providing it as part of an open API to other government systems and pipelines. This system will build on LCC’s cross-lingual knowledge base Lorify (Monahan and Carpenter 2012), as well as LCC’s customization systems developed for the Air Force.  These systems not only support low-shot or weakly supervision customization of the system, but also perform unsupervised analysis of relevant sources to identify potential candidates for customization, both speeding up the customization process and supporting analysts in quickly identifying the relevant concepts that require customization.