Advancement of Targetable NLP

In this project we are developing a system that can be rapidly trained to create a customized NLP data extraction pipeline. The pipeline can extract and aggregate information about new entities and relations as defined by an end user. ĀThe approach, called TELEx (Type Enhanced Learning for Extraction), builds on recent progress on Large Language Models and few-shot extraction of fine-grained entities.Ā Our technology enables end users to easily and rapidly define to new entities and relations, and train a machine learning component to extract these entities and relations.Ā In addition, we have incorporated TELEx in an end-to-end text processing and analysis platform, DDEx, that has capabilities for data harvesting, extraction, data cleaning, entity resolution, information integration, inference and search. With the additional of TELEx, the DDEx platform can now be rapidly customized to meet the domain-specific needs of customers. The work in this proposed SequentialĀPhase 2 project focuses on expanding and improving DDEx/TELEx so that it can be more flexibly integrated with larger external systems, including specifically systems currently being developed by the ARMY for intelligence and related purposes. The improvements are designed to support a variety of options for configuring, packaging and integrating DDExÆs functionality in a larger environment. This includes improvements to the systemÆs modularity and configurability, as well as extending the range of the systemÆs capabilities for reformatting, cleaning, and linking data. ĀIn addition, further enhancements will be made to ensure that the systemÆs performance meets the requirements of the Army and other customers.