On the Influence of Models-to-Natural-Language Transformation in Traceability Link Recovery among Requirements and Conceptual Models

Recovering traceability links between software artifacts and requirements is a common task in Software Engineering. Information Retrieval (IR) techniques have been applied to recover traceability links amongst code and requirements. By transforming Models into Natural Language (M2NL), it is possible to apply IR to calculate their traceability links to requirements. However, results retrieved by IR are affected by the writing style of the NL input. Regarding M2NL, there are two main types of techniques in use: Rule-Based techniques, and Element Based techniques. Along with M2NL, there is a wide range of Natural Language Processing (NLP) techniques that can be applied. Through this work, we analyze how the usage of distinct M2NL-NLP combinations of techniques impacts IR-based Traceability Links Recovery over requirements and models. We evaluate two different M2NL techniques, and the inclusion of Simple and Advanced NLP along with M2NL, in a real-world industrial case study.