Automated Detection of Language Issues Affecting Accuracy, Ambiguity and Verifiability in Software Requirements Written in Natural Language

Most embedded systems for the avionics industry are considered safety critical systems; as a result, strict software development standards exist to ensure critical software is built with the highest quality possible. One of such standards, DO-178B, establishes a number of properties that software requirements must satisfy including: accuracy, non-ambiguity and verifiability. From a language perspective, it is possible to automate the analysis of software requirements to determine whether or not they satisfy some quality properties. This work suggests a bounded definition for three properties (accuracy, non-ambiguity and verifiability) considering the main characteristics that software requirements must exhibit to satisfy those objectives. A software prototype that combines natural language processing (NLP) techniques and specialized dictionaries was built to examine software requirements written in English with the goal of identifying whether or not they satisfy the desired properties. Preliminary results are presented showing how the tool effectively identifies critical issues that are normally ignored by human reviewers.