https://www.mdu.se/

A clear understanding of customers' requirements is fundamental towards developing products that behave as intended. Customers commonly use natural language (NL) to specify their requirements. As NL is innately ambiguous and an industrial project could contain thousands of specifications., requirement analysis becomes a highly demanding and time-consuming task. One of the goals in industry is, therefore., to minimise the amount of time spent manually analysing requirements. This article presents a natural language processing (NLP) approach to automatically classify rail domain requirements based on the presence of ambiguity., and to provide textual explanations regarding the reason behind the classification. Traditional machine learning (ML) classification models are trained using lexical features from requirements and experts' comments concerning ambiguity on annotated real-world data. 10-fold cross-validation results show an F-score up to 0.87., with a recall up to 0.88. Furthermore., a validation of the model with 100 additional requirements achieved an accuracy of 0.78., with 76% match between the model's and expert's classification. The provided explanations are important for the expert in understanding the key decision terms involved in the classification., as well as provided insights on the presence of ambiguities in requirements. Ours is among the first works that uses explainability to tackle ambiguity in textual requirements., employing NLP and ML.