Automatic Detection and Resolution of Lexical Ambiguity in Process Models

Abstract

Process models play an important role in various system-related management activities including requirements elicitation, domain analysis, software design as well as documentation of databases, business processes, and software systems. However, it has been found that the correct and meaningful usage of process models appears to be achallenge in practical settings requiring the usage of automatic model analysis techniques. Up until now, such automatic quality assurance is mainly available for checking formal properties of process models. Forinstance, there is arich set of analysis techniques to check control-flow-related properties of process models, such as soundness. There are only afew techniques available for checking guidelines on text labels with regard to terminological ambiguity.Moreover, the terminology problem is even more serious in process models since the process model givesa na bstract viewo ft he business process and provides only limited context to detect and resolvea mbiguity issues. It is thus the goal of [PLM15] to address the need for automatic techniques as well as to define detection and resolution technique for textual ambiguities that improve the terminological quality of aprocess models and repositories thereof.Fort hat purpose, our approach addresses three important issues, i.e. the manual effort, the missing focus on process model text fragments, and the focus on single models. First, the required manual effort,refers to the extensive amount of manual work that is required to detect and resolvea mbiguities in process models. The human effort can be tremendous since organizations tend to maintain several hundreds or even thousands of process models. Second, the missing focus on process model text fragments refers to the fact that manyapproaches of ambiguity detection and resolution are tailored to deal with sentences and phrases taken from ag rammatically and syntactically correct natural language text. However, the elements of process models contain only short textual fragments that do not exhibit ac omplete or ac orrect sentence structure impeding the direct application of such approaches. Third, the focus on single models relates to the observation that available techniques consider only single models or smaller units thereof. Hence, these techniques address ambiguities within as ingle document or process model. However, since we assume ar epository of several process models, the correction of ambiguities on document levelmight introduce an inconsistencyinanother document or model.Our approach introduces the notion of semantic vectors that represents all the possible meanings of at erm in the context of ap rocess model. As emantic vector interprets the

Comparing and Aligning Process Representations

“…These approaches analyze the quality and consistency of activity labels in a model, for example by detecting and/or correcting inconsistent use of terminology [148] or violations of labeling conventions [45,164,271]. Others aim to improve modeling quality by detecting common modeling errors [120] or ambiguously labeled activities [213,214]. Some approaches add information to a process model by annotating model elements with semantic or ontological information [49,97,165].…”

From Process Models to Chatbots

“…For instance, the label Retrieve parts from storage in swimlane Department in Figure 1 would produce the semantic structure in Figure 4. We use a custom grammar and not a general purpose natural language parser such as those provided by FreeLing or other similar library because of the particular structure of model task labels: Task labels are commonly written in simple patterns action-object (retrieve parts), or object-nominalized action (parts retrieval ) with sometimes some additional complement(s) [11]. Also, the subject is usually ommitted, which causes general purpose PoS taggers and parsers to fail more often.…”

Searching textual and model-based process descriptions based on a unified data format

“…Van der Vos et al [51] use a semantic lexicon to check whether the terms used in a model are sufficiently meaningful. Pittke et al [43] take a more general perspective and, among others, check whether process models contain misleading synonyms. Based on such semantic considerations, also techniques for detecting errors in process models have been defined.…”