Presentation
Semi-Automated Development of Textual Requirements: Combined Natural Language Processing and Multi-Domain Semantic Approach
Publication Date: 9/21/2022Start Date: 2022-09-21
End Date: 2022-09-22
Event: AI4SE and SE4AI Workshop 2021
Event: Stevens Institute of Technology, Howe Building, Hoboken, NJ
Lead Authors:
Dr. Mark Austin
Abstract
Modern engineering systems are nearly always designed, built, operated, and maintained by teams of people and automated procedures for decision making over extended periods of time. A key element in bringing this diversity of human and automation capability together and making systems that work (avoid unnecessary losses and failures) is the ability to write and manage textual requirements early in the system development life cycle. And despite remarkable advances in AI/ML over the past twenty years, use textual requirements for contractual, regulatory and legal agreements persists.
In a step toward addressing these challenges, we present a framework for the semi- automated development of and validation of textual requirements. The proposed framework is novel in three respects. First, it employs a combination of natural language processing and template matching to restrict the range of ways in which textual requirements can be stated. Clearly written requirements improve the quality of communication among disciplines throughout the system life cycle—they are a key factor in the reduction of uncertainties and improvements to project schedule. Second, knowledge-based models work hand in hand with NLP and templates to help designers fill in the details of a model that might be missing or remain to be specified. Third, we envision the proposed approach being part of a multi-domain semantic modeling and reasoning framework that includes the relevant engineering domains and their connections to processes for project management and governance. In a departure from state- of-the-art approaches to semantic model development, which tend to focus on comprehensive descriptions of knowledge for a particular domain, our approach puts data models, ontologies, and rules on an equal footing and provides computational support for the executable processing of incoming events. The proposed method employs a judicious combination of domain-neutral and domain-specific ontologies and rules. We use the former for the semantic modeling of time units and physical quantity units, and extend basic usage to cover the verification and validation of individual and groups of related textual requirements.
A software prototype called FLOOR has been developed for the high-level system integration of requirements, templates, ontologies, and rules. We will describe these capabilities for textual requirement sentences covering project cost and schedule, building foundation, construction site, equipment, and workers, inspired by the (recently constructed) Brendan Iribe Center for Computer Science and Engineering at the University of Maryland in College Park.