Learning Environmental Assumptions via Natural Language Processing (NLP)
What went wrong in Ariane 5 rocket’s launch? A software error, but not a bug in the source code implementation, rather a flawed environmental assumption: The assumed maximum horizontal velocity based on the previous system (Ariane 4) did not hold for Ariane 5. As a result, an overflow error occurred and both the primary and backup guidance components failed. Unfortunately, the Ariane 5 accident was only one of the many system failures arising from missing or invalid environmental assumptions. This Protégé research project is aimed at using natural language processing (NLP) techniques and artificial intelligence to learn environmental assumptions from online human knowledge sources (e.g., Wikipedia), and further use these assumptions to carry out requirements-based testing activities in order to uncover defects and errors. The research will investigate NLP methods (e.g., dependency parsing) to discover linguistic constructs, and then use unsupervised learning and generative models to produce relevant and useful conditions about the environment in which a software-intensive system operates. The Protégé student will join a team of graduate students in the Software Engineering Research Lab to develop novel and scalable ways to improve software quality.
Associate Professor, CEAS - Computing Sciences & Informatics
832 Rhodes Hall