Trial lecture
"Language Technologies for Code Analysis”
Time and place: October 11, 2024 11:15 AM. Kristen Nygaards sal (5370), Ole Johan Dahls hus.
Main research findings
Requirements describe qualities that artefacts or services must have and are important drivers in many industries. When the number of requirements grows, managing requirements as text documents becomes difficult. Using machine-readable requirements (e.g., logical formulas) can enable increased automation.
We explore several natural language processing (NLP) strategies to help transform existing requirements into a machine-readable format (semantic parsing). Since requirement documents are written for experts and training data is rare, using supervised techniques based on neural networks is non-trivial as they require thousands of training examples.
We demonstrate that it is possible to use openly available data to help create training data for text classification of requirements. Further, we show that identifying key components in a requirement sentence can be seen as a sequence labelling task, where each token is assigned a label. We argue that the requirements are very context-dependent and show that making the document-level context explicit helps detect these components.
We present an algorithm for human-machine collaborative annotation, where a large language model assists in annotating requirements sentences with description logic formulas. The findings in the thesis can reduce ambiguity and make it easier to identify relevant requirements which can improve safety.
Adjudication committee
- Professor Melinda Hodkiewicz, The University of Western Australia, Australia
- Professor Andreas Vogelsang, University of Cologne, Germany
- Professor Adin Ramirez Rivera, Department of Informatics, University of Oslo, Norway
Supervisors
-
Researcher Basil Fabian Ell, Department of Informatics, University of Oslo, Norway
-
Professor Martin Giese, Department of Informatics, University of Oslo, Norway
-
Professor Lilja Øvrelid, Department of Informatics, University of Oslo, Norway
Chair of defence
Professor Jens Johan Kaasbøll, Department of Informatics, University of Oslo
Contact information at Department: Mozhdeh Sheibani Harat