Requirement Phrasing Assistance Using Automatic Quality Assessment

The design of modern hardware systems is a very complex and time consuming process. At the beginning of the design process, requirements need to be specified. Errors in that early design stage derived by misinterpretation of the requirements can be hard to detect and require significant effort and costs to get fixed. To prevent errors, requirements should be written in a comprehensive and unambiguous way. Thus, designers are interested in automatic assistance tools that help writing better requirements. Conventional approaches are usually rule-based, thus many syntactic and semantic properties are not considered. In this paper we introduce an alternative approach to ensure the quality of requirements. The approach has two stages and assists the designer by providing (i) all relevant statistics about the syntactic and semantic properties of the sentence, and (ii) a single consolidated nominal quality predicate for the sentence such as good, medium, or bad. Although such statistical quality assessment leads to already satisfying results, the algorithm prediction reliability can further be enhanced by machine learning techniques. The achieved reliability for quality assessment in combination with the overview of the metrics about syntax and semantic can help the designer to write more comprehensive and less ambiguous requirements.