Semantic Frames for Classifying Temporal Requirements: An Exploratory Study

Temporal requirements express the time-related system behaviors and properties. In engineering critical systems, experience has shown that temporal requirements are the most problematic type of requirements to verify. Researchers have thus used natural language processing (NLP) techniques—most notably, part-of-speech (PoS) tagging—to develop practical classifiers to distinguish temporal requirements from non-temporal ones. In this paper, we explore frame semantics—a linguistic approach to labeling a word’s role in a sentence with respect to the events of interest—to augment the temporal requirements classification task. Our experiments of 111 requirements sentences from the regulatory documents show that the best classification accuracy of 90.9% is achieved when PoS features are replaced with, rather than combined with, frame semantics features. The results suggest the promising role of semantics-augmented NLP support in an understudied requirements engineering task.