Automated Quality Assessment of Natural Language Requirements

High demands on quality and increasing complexity are major challenges in the development of industrial software in general. The development of automotive software in particular is subject to additional safety, security, and legal demands. In such software projects, the specification of requirements is the first concrete output of the development process and usually the basis for communication between manufacturers and development partners. The quality of this output is therefore decisive for the success of a software development project. In recent years, many efforts in academia and practice have been targeted towards securing and improving the quality of requirement specifications. Early improvement approaches concentrated on the assistance of developers in formulating their requirements. Other approaches focus on the use of formal methods; but despite several advantages, these are not widely applied in practice today. Most software requirements today are informal and still specified in natural language. Current and previous research mainly focuses on quality characteristics agreed upon by the software engineering community. They are described in the standard ISO/IEC/IEEE 29148:2011, which offers nine essential characteristics for requirements quality. Several approaches focus additionally on measurable indicators that can be derived from text. More recent publications target the automated analysis of requirements by assessing their quality characteristics and by utilizing methods from natural language processing and techniques from machine learning. This thesis focuses in particular on the reliability and accuracy in the assessment of requirements and addresses the relationships between textual indicators and quality characteristics as defined by global standards. In addition, an automated quality assessment of natural language requirements is implemented by using machine learning techniques. For this purpose, labeled data is captured through assessment sessions. In these sessions, experts from the automotive industry manually assess the quality characteristics of natural language requirements.% as defined in ISO 29148. The research is carried out in cooperation with an international engineering and consulting company and enables us to access requirements from automotive software development projects of safety and comfort functions. We demonstrate the applicability of our approach for real requirements and present promising results for an industry-wide application