Ambiguity Detection in Business Process Descriptions: An Evidence and an Automated Approach

This paper examines the role of ambiguity in process descriptions and how feasible it is to identify ambiguities using automated techniques. Process descriptions are natural language specifications of a business process. An ambiguity occurs whenever part of the specification can be understood in multiple ways. Ambiguities harm model quality, as well as comprehension and business process execution. This paper provides two contributions. First, it examines the prevalence of ambiguities in process descriptions. Second, it explores how the creation of computational artifacts could support the identification of ambiguities at design time. We constructed a dataset of process descriptions considering multiple authors, writing styles, and possible uses in academia and industry. In a dataset of N=71 process descriptions, we show that ambiguities are a prevalent phenomenon, where 94,3% of the descriptions exhibit at least one type of ambiguity. Concerning identification, we can see that standard classifier techniques fall short in identifying ambiguities. At the same time, the application of pre-trained and fine-tuned LLMs improves the identification of ambiguity significantly. By demonstrating the presence of ambiguities in process descriptions and identifying them, our work aims to influence research on the application of Natural Language Processing (NLP) in Business Process Management (BPM) by refining pipelines to consider multiple interpretations. Moreover, it advocates for the transformative role of NLP as a tool that can promote clearer and more reliable process definitions.