Ontological Analysis Approach

In this section we discuss the ontological analysis approach we use. We apply ontological analysis in chapters 5 and 6.

Ontological analysis is performed by considering a mapping between language modeling constructs and the concepts in an ontology. In ontological analysis, each modeling element is interpreted using the ontological theory as a semantic domain, i.e., the meaning of each modeling element is interpreted based on the concepts of the ontological theory. In this work, we make use of our ontological basis to propose our theoretical foundation, which is described in chapters 5 and 6. Our theoretical foundation is then used for ontological analysis purposes as an ontological theory. The performance of ontological analysis further stresses the relevance of the theoretical foundation, as well as the relevance of its correctness and expressiveness¹⁵.

Further, concepts of the domain of discourse (captured in the theoretical foundation) should be represented by modeling elements of the language being considered (Weber, 1997) (Moody, 2009). When this is not the case, the modeler is not able to represent some part of the domain of discourse. According to (Weber, 1997), there should be a one-to-one correspondence between the concepts in the ontology and the language modeling elements. When this correspondence cannot be obtained, the following language problems can be identified¹⁶, as illustrated in Figure 11:

15 The ontological analysis is not used to enforce any claim on the theoretical foundation correctness. However, we disclaim that correctness and expressiveness is relevant in the theoretical foundation since the theoretical foundation is used in this thesis for ontological analysis, serving as an ontological theory.

16 Adapted from (Santos Jr et al., 2013)

Figure 11 - Issues uncovered by ontological analysis - adapted from (Moody, 2009)

• Construct excess: exists when a modeling construct does not correspond to any ontological concept. Since no mapping is defined for the exceeding construct, its meaning becomes uncertain, hence, undermining the clarity of the specification produced using the language.

According to (Wand and Weber, 1990), users of a modeling language must be able to make a clear link between a modeling construct and its interpretation in terms of domain concepts. Otherwise they will be unable to articulate precisely the meaning of the unclear construct and, consequently, the specifications¹⁷ they generate using the language.

Therefore, a modeling language should not contain construct excess and every instance of its modeling constructs must represent an individual in the domain.

• Construct overload: exists when a single modeling construct can represent multiple ontological concepts. Construct overload impacts language clarity negatively. Construct overload is considered as an undesirable property of a modeling language since it causes ambiguity and, hence, undermines clarity. When a construct overload exists, users have to bring additional knowledge not contained in the specification to understand the phenomena which are being represented.

• Construct redundancy: exists when multiple modeling constructs can be used to represent a single ontological concept. Construct redundancy is a

17 Specification is used as specified model, the user defined model.

violation of parsimony. In (Wand and Weber, 1990), the authors claims that construct redundancy “adds unnecessarily to the complexity of the modeling language” and that “unless users have in-depth knowledge of the grammar, they may be confused by the redundant construct. They might assume for example that the construct somehow stands for some other type of phenomenon.” Therefore, construct redundancy can also be considered to undermine representation clarity.

• Construct deficit: exists when there is no construct in the modeling language that corresponds to a particular ontological concept. Construct deficit entails lack of expressivity of the modeling language, i.e., there are relevant phenomena in the considered domain (according to a domain conceptualization) that cannot be represented by the language.

Alternatively, users of the language can choose to overload an existing construct, thus, undermining clarity.

In this work, we argue that a one-to-one mapping is too restrictive and agree with (Carvalho, Almeida and Guizzardi, 2014), which argues that patterns of modeling elements in the language could also be considered for the mapping correspondence. Some concepts in the domain of discourse might be too complex to be covered by a single modeling element. In this case, these domain of discourse concepts might be represented by a pattern of modeling elements in the modeling language. In this sense, patterns of elements (i.e., a subset of language elements that are to be used together) should also be target for ontological analysis. Further, we also account that, for the sake of simplicity, it might be of the interest of language creators that a single modeling element might represent a pattern on the domain of discourse. We also analyze language modeling elements with respect to patterns of concepts in the domain of discourse.

To conclude, we argue that modeling languages aiming at wide adoption, such as ArchiMate, which is analyzed in the upcoming chapters, should avoid ontological deficiencies. Evidences from empirical studies reported that users of modeling languages perceive when ontological deficiencies exist and that these deficiency perceptions are negatively associated with usefulness and ease of use of languages (Recker et al., 2011). In the next chapter we discuss strategic planning and propose a conceptual model for representing strategic plans in line with both the management literature and with the ontological analysis method described in this section.

Chapter

5