Natural Language Processing in Requirements Engineering and Its Challenges for Requirements Modelling in the Engineering Design Domain

Abstract

Requirements represent a central element in product development. The large number of requirements inevitably results in an increased susceptibility to errors, an expenditure of time and development costs. The associated problems motivate the application of Artificial Intelligence in the form of Natural Language Processing (NLP). In Requirements Engineering one main task is the classification of requirements which serves as the input in architectural models e.g. in SysML. In mechanical engineering there is still little overview regarding the interface between requirements classification and modelling. This paper provides an overview of the requirement classes and entities used in the literature and analyses their utilisation in modelling. Existing requirements classes usually do not offer the flexibility to be transferred to other domains. However, basic structures can be adopted from those classifications. This enables a clear assignment of existing classes to object classes in modelling. Resulting from the conducted literature study the observed predominant focus of research on the software industry requires an extension of the existing requirement classes and entities to enable further use and transfer to mechanical engineering.

Automatic derivation of use case diagrams from interrelated natural language requirements

“…Nevertheless, strict adherence to these rules is not always observed, and variations are explicitly accepted as long as the content is unambiguously conveyed. This potentially leads to a heterogeneous set of requirements (van Remmen et al, 2023). That implies that rule-based formalizations demand a significant number of rules, thus resulting in a considerable amount of manual labour initially and in case of changing requirement styles.…”

A comparative experiment between textual requirements and model‐based requirements on proxies for contractual safety

“…6 In fact, imprecision seems to be inherent to natural language, leading to requirements expressed in natural language often being ambiguous, incomplete, and inaccurate, [7][8][9][10] as well as prone to manual errors. 11 In addition, natural language is difficult (often impossible) to parse into consistent logical or mathematical statementsm, [12][13][14] which limits the use of systematic and/or automated tools to explore completeness. 2,15,16 Model-based requirements have been proposed as an alternative to textual requirements, with the promise of enabling higher accuracy, precision, and completeness when eliciting requirements.…”