Rapid Acquisition of Hierarchical Procedures from Instructional Documents

Abstract

Many Navy activities involve the execution of complex procedures that would benefit from automation or assistance. Artificial intelligence techniques can help satisfy this need, but they depend on domain expertise that often cannot effectively be entered manually or learned from large data sets. Instead, we propose to automate the extraction of knowledge about procedures from manuals and related documents. We will encode activities as hierarchical task networks, a modular and interpretable formalism that supports learning. The project will build on three existing systems START, Genesis, and ICARUS that have complementary abilities in syntactic processing, semantic interpretation, and incremental procedure learning. We will combine these into an integrated computational system that parses sentences from instructional text, interprets their meaning using background knowledge, and invokes abductive learning to transform the meaning structures into hierarchical procedures. The team will obtain instructional documents for three domains mealpreparation, shipboard equipment repairh, and aircraft maintenance which it will use demonstrate and evaluate the systems abilities.

Document Details

Document Type
DoD Grant Award
Publication Date
Jun 17, 2020
Source ID
N000142012643

Entities

People

  • Patrick Langley

Organizations

  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Computational Linguistics
  • Database Systems and Applications
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Information Retrieval