Extracting Traceable Formal Models from Natural Language Policy Documents

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

0429948 Insup Lee University of Pennsylvania Extracting Traceable Formal Models from Natural Language Policy Documents Insup Lee, Aravind Joshi Policy plays an important role in our lives and affects us in many ways, e.g., the Food and Drug Administration's Code of Federal Regulations govern how to test blood for communicable diseases. Ambiguities, conflicts, and incompleteness in such policy documents could lead to situations that are undesirable and unsafe. The proposed research is to develop NLP (Natural Language Processing) based techniques and methods for extracting formal models from policy documents. These models are then analyzed for correctness and consistency and also to used for conformance testing of implementations of the policy. This is a collaborative effort between researchers in NLP and Formal Methods and aims at producing an environment in which policy can co-exist in natural and formal languages. For success and usefulness of this approach, it is important to maintain correspondence and traceability between these two representations of policy. Furthermore, the large size of the policy bases and the complexity of the documents warrant modularized extraction of models and then the merging of these models. Existing NLP techniques need to be extended and tailored to aid in the modular extraction of formal models. The merging of extracted models also requires extensions and refinements to formal method techniques. As our society relies more on computer-based systems, and on medical devices in particular, the proposed research will help to improve the reliability of such systems.