On Linguistic Quality of Natural Language Requirements

We recognize the need for deening a quality model of linguistic properties of natural language requirements. We are convinced that the deenition of a thorough linguistic quality model can be obtained as a meeting point between the specialization of high level quality frameworks for requirements engineering on one hand, and a systematic collection and classiication of the diierent Natural Language Processing techniques proposed in the literature speciically for the analysis of requirements. 1 Motivations Quality in software engineering is following, to some extent, the same historical path as quality in classical engineering has. Many methods and tools have been proposed in the past to achieve some speciic quality objectives. However, the vast majority of them are the outcome of a painful trial-and-error process. Stories of projects ended in uttermost disaster have taught us what not to do in a software project, while experiences with successful projects, together with some common sense, have taught us what to do. Such an approach, although useful during the childhood of our discipline, has now reached its limits. Like classical engineering switched from quality control to total quality management, we need to switch from \best practices" to a systematic, comprehensive, and unifying approach in understanding and analyzing problems. In this paper, we concentrate on quality of natural language (NL) requirements. Natural language properties deserve a particular attention since the main purpose of a requirements document is to convey information, gathered from the customer and other sources, to the developers, and this communication is mediated by the language used. In recent years, several attempts have been conducted in order to apply general Natural Language Processing (NLP) techniques to the purpose of analyzing and controlling software requirements. As an example, some works (e.g., 12]) investigated how the construction of a conceptual model of a problem is related to the linguistic mechanisms through which analysts abstract natural language requirements onto concepts. Unfortunately, the eeectiveness of the proposed techniques and tools in aaecting quality of requirements has not been demonstrated in most cases. The reason lies in the fact that no systematic