Nuts and Bolts of Extracting Variability Models from Natural Language Requirements Documents

Abstract

Natural language (NL) requirements documents are often ambiguous, and this is considered as a source of problems in the later interpretation of requirements. Ambiguity detection tools have been developed with the objective of improving the quality of requirement documents. However, defects as vagueness, optionality, weakness and multiplicity at requirements level can in some cases give an indication of possible variability, either in design and in implementation choices or configurability decisions. Variability information is actually the seed of the software engineering development practice aiming at building families of related systems, known as software product lines. Building on the results of previous analyses conducted on large and real word requirement documents, with QuARS NL analysis tool, we provide here a classification of the forms of ambiguity that indicate variation points, and we illustrate the practical aspects of the approach by means of a simple running example. To provide a more complete description of a line of software products, it is necessary to extrapolate, in addition to variability, also the common elements. To this end we propose here to take advantage of the capabilities of the REGICE tool to extract and cluster the glossary terms from the requirement documents. In summary, we introduce the combined application of two different NL processing tools to extract features and variability and use them to model a software product line.