Scalable Prosodic, Anomaly and Relational Knowledge Exploration of Language with Enhanced Robustness (SPARKLER)
Abstract
We have developed trilingual (English, Chinese and Spanish) open-domain knowledge extraction techniques to extract, populate, and analyze unstructured data from heterogeneous sources in to acknowledge base, to relieve the reliance on manual knowledge base construction or in-person communication for knowledge management. This new frame work is able to discover schemes and extract facts from any input data in any domain, without any annotated training data, by incorporating distributional semantics and symbolic semantics. The resulting systems have won top performance at several NIST international research evaluations and been selected for DARPA and ARL demos and transitions. 4 PhD students have been supported, and 54 papers have been published at top conferences and journals.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2018
- Accession Number
- AD1052273
Entities
People
- Andrew Rosenberg
- Heng Ji
- Yang Liu
Organizations
- Research Foundation of The City University of New York