Automatic Extraction and Formalization of Temporal Requirements from Text: A Survey

Natural Language Processing has opened new paths for business process management and requirements engineering, particularly in automating the extraction and formalization of temporal requirements from diverse documents such as system specifications, legal texts, and business process descriptions. Recently, approaches have been introduced to automate this task, employing various document formats as input and targeting different formal specifications. However, a key challenge persists: effectively comparing these approaches and choosing the most suitable one for a specific task. This paper aims to bridge this research gap by conducting a systematic literature review, including a detailed analysis and comparing existing approaches. This comparison is crucial to determine if the latest Large Language Model-based solutions could surpass existing methods in effectiveness and ease of use. The systematic literature review enables users to select the most suitable method based on their data and end goals. Moreover, this work proposes the NL2MTL (https://github.com/marisol-barrientos/nl2mtl, DLA: 22.04.2024) method to bridge some of the gaps identified in the literature analysis, i.e., establishing a comparable assessment method, under-representation of legal texts, poor output context management, and the necessity to automate the formalization of requirements, considering both quantitative and qualitative aspects of time. Addressing the latter aspect, we select Metric Temporal Logic (MTL) as formalization and provide the associated prompts and an evaluation of the NL2MTL output.