Related Work

We find one of the early discussions of powertypes in the paper by Odell [Odel1994]. Also, Guizzardi et al. discuss “What’s in a Powertype?” [GuAl2015]. The article by Carvalho and Almeida [CaAl2016] describes how the powertype MobilePhoneModel can be modeled using so-called regularity attributes. However, in their approach, the classes MobilePhone and MobilePhoneModel cannot be merged, because then each physical phone would also have a launchDate. Atkinson and Kühne [AtKu2001] argue that the shallow instantiation approach cannot solve the problems of multiple classification and replication of concepts. The authors add the feature potency to each model element. In each instantiation step, the potency is reduced by 1. However, this requires that the entities to be modeled so that they can be divided into logical levels, each of which must be assigned a meaning. This introduces the potential for layer mistakes, especially in ontologies that change frequently during development. The subdivision of base classes is also often done using terms such as type, category, family, group, order, subordinate, and so on. For example, Neumayr et al. ([NeSc2008] on page 9, in Figure 3), model the base classes Book and Car as subclasses of ProductCategory and assign it a classification level of 3. To model that marketLaunch must be instantiated at the brand level and mileage must be be instantiated by objects at the physical entity level, the authors claim on page 22 that it is necessary to use explicit constraints. Since the authors do not specify the constraints, the question of how this can be achieved is left open. We also see the potential for layer mistakes in taking this approach. Guizzardi et al. [GuAl2015] also discuss deep instantiation. The powertype class BirdSpecies and the base class Bird are modelled in two parallel branches of a class hierarchy. We identify a modeling deficit in which the properties of all other species should have to be modeled redundantly as in BirdSpecies, or should be moved to a higher class Species. We consider the graph rewriting rules on page 6 of [GuFi2019] to be an important contribution to reducing the complexity of ontology models. Another meta-level model for the biological domain by Batista et al. [BaAl2022] exemplifies categorization and the use of powertypes. Again, here the question arises, whether the set of rules and axioms for creating multi-level structures listed on page 18 could be reduced by an alternative modeling method. The approach of Bense and Humm [BeHu2021] proposes to satisfy the instance and type character by two different inheritance mechanisms of properties. These are BroderInheritedProperty based on skos:broader and TypeInheritedProperty based on rdfs:class. A case not covered by this is when a modeler wants to express that all polar bears have white fur. Bense already discusses in [Bens2023a] three different types of data properties for Meta-Level Modeling (MLM). In comparison, we show in our paper that it is necessary to introduce a fourth type called Transparent Data Property (TDP). Fonseca et al. [FoAl2021] provide a summary of multi-level approaches and their capabilities according to a list of eight multi-level features. They also discuss MLM features such as durability, mutability and potency, e.g., ‘In DeepJava, the potency of an element denotes the maximum depth of its instantiation chain, or how many times a type can be instantiated’ or ‘However, the only mechanism available for relating features across levels is potency, which is limited to defining how deep a feature is present in an instantiation chain’. In our study we provide an easier to handle method compared with using potency. Attributes that can be instantiated only once in an instantiation chain are considered to be single-potency elements. The authors also discuss the ‘multi-level phenomenon called deep instantiation, where the attributes of a higher-order type affect entities at lower levels’. As an example, they mention that when lions are assigned the feature warmblooded = T, this also applies to all particular lions. For the desired functionality that an attribute-value assignments in a class propagates down an instantiation chain to all of the particular instances of those classes we will introduce the so-called datatype Transparent Data Property (TDP). Within the Dublin Core Metadata Initiative, metadata is defined in the Dublin Core Metadata Element Set. Are data properties like Title, Publisher and Description really always metadata, or are they application domain specific? It needs to be clarified how to achieve flexibility in modeling so that such attributes can be used to represent either knowledge domain specific information or meta information about the domain model, i.e., information about the model of the domain model.