Motivation
Models are important to manage complexity. They provide a means for understanding the business process, and understanding already is a benefit. This is indicated by a study from Gartner revealing an increase in efficiency of 12 percent gained solely by documenting actions and organizational responsibilities using process models (Melenovsky 2005, p. 4). Moreover, process models serve for optimization, reengineering, and implementation of supporting IT systems. Due to the importance of process models, model quality is important. According to ISO 8402, quality is “the totality of characteristics of an entity that bear on its ability to satisfy stated and implied needs”. Facets of quality are – amongst others – adequate coverage of the domain or system to be modeled, appropriateness in respect to the abstraction level of the representation (scale), detail of representation (granularity) and the correctness of a model. We concentrate on correctness as the most fundamental quality aspect. Among the aspects of correctness are: (a) syntactical correctness, (b) correctness in regard to the formal semantics, (c) correctness in regard to linguistic aspects focusing on the labels used in models, (d) correctness in regard to the coherence of connected models and (e) compliance to rules and regulations focusing on the correctness of the model’s content and thus on semantic correctness. While there are numerous verification approaches available to ensure (a-d), only a few approaches focus on (e) in the sense of the verification of the semantic correctness. With the term “verification”, we denote criteria targeting the internal, syntactic and semantic constitution of a model. In contrast to that, validation means the eligibility of a model in respect to its intended use (Desel 2002, p. 24) – in other words: if the criteria is something outside the model (Chapurlat & Braesch 2008; Mendling 2009, p. 2). Following this distinction, we call the procedures to ensure semantic correctness “semantic verification”.
A major problem regarding semantic verification is how to automate it. This problem is rooted in natural language being used for labeling model elements, thus introducing terminological problems such as ambiguity (homonyms, synonyms) and other linguistic phenomena. Model creators and readers do not necessarily share the same understanding as the concepts they use are usually not documented and mix both discipline-specific terminology and informal, ordinary language. Therefore, it is hard for humans to judge if a model is semantically correct and almost impossible for machines (apart from using heuristics) because the model element labels are not backed with machine processable semantics. The result is that the machine cannot interpret the contents of model elements. Our solution approach is to encode the model element semantics in a precise, machine readable form using ontologies. Further, we then use rules to encode constraints used for verifying aspects of semantic correctness.