Related Work

The paper by Guizzardi et. al [GuFi2019] discusses various aspects of partitioning. They mention ‘subkinds’, ‘kinds’, and ‘phases’ as varieties of ‘dividing principles’. Phase partitioning is characterized as dynamic partitioning, i.e., if we classify persons according to their ‘life phase’ it can change from ‘baby’ to ‘child’ to ‘adolescent’ to ‘adult’ to ‘deceased’, and so on. But a person can belong only to precisely one ‘life phase’ at any one a time. Thus, ‘life phase’ can additionally be characterized as a case of single-valued partitioning. An example of static partitioning would be whether a ‘biological organism’ has the property ‘warm-blooded’ or not. The authors define subkind partitions as being static, i.e., in their view an entity cannot move from one subkind partition to another, i.e., subkinds are not mutable. As an example they mention ‘Car Agency’ to be a static subkind of ‘Organization’. Furthermore, the paper describes enumeration datatypes as an abstraction method for representing subkind and phase partitions. This is a very practical approach and the visualization using an UML diagram is easy to understand. On the other hand, it does not solve the problem of how to maintain additional metadata such as the cardinality of entities of a partition.

Through the idea of dynamic classification Carvalho and Almeida [CaAl2016], page 29, try to support the idea that both individuals and types can change qualitatively, while maintaining their identity. This corresponds to the dynamic partitioning mentioned in [GuFi2019]. We will discuss how both aspects can be supported by the feature mutability. In addition, in [CaAl2016], page 17 the definition D6 is used to describe the complete categorization, and as an example they declare that “EmployeeAcademicDegreeType” partitions “Employee”. The authors characterize “EmployeeAcademicDegreeType” as an instance of “2ndOT” (2nd Order Type). They also introduce subordination between the higher-order types t1 and t2 and so-called cross-level structural relations and, they also introduce proper specialization as a spezialization where properSpecializes(t1, t2) adds the constraint that at least one instance of t2 is not an instance of t1. Their theorem T18, page 18, captures the case: “A consequence of the partitions definition is that, if two types t1 and t2 both partition the same type t3 then it is not possible for t1 to specialize t2”. For example, this holds for t3=¬feathered-warm-blooded, t1=¬feathered, t2=warm-blooded. We will examine how partitions (t1,t2) can be generalized with respect to the combination of n>2 types. The paper [CaAl2016], page 14, also describes how a powertype MobilePhoneModel can be modeled using regularity attributes that determine how MobilePhone can be partitioned by its powertype MobilePhoneModel. However, the MobilePhone and MobilePhoneModel classes cannot actually be merged using regularity attributes, since each particular phone would then also have a launchDate. I.e., they cannot streamline there modeling by benefitting from merging pairs of powertypes with their base types. We think that subsuming the powertype under its base type is desirable since both model aspects of the same concept. This raises the question of how this could be accomplished in order to streamline the modeling. The work cited so far uses additional information, especially for graphic visualization in UML-like diagrams that is not directly stored as meta-information when the ontologies are saved. This raises the question regarding how this deficit can be avoided by an alternative modeling method and how this method has to be general enough to not only of single properties but also combinations of properties.