Source-linked AI summary
Category-Based MLM: Unifying Powertypes with Superclasses
Mira Balaban, Azzam Maraee, Arnon Sturm
TL;DR
MLM uses interlevel instance-of relations to model multiple classification levels, but deep characterization gives category classes a superclass-like influence that membership alone does not explain. CatMLM divides category and object features to give category classes powertype and superclass facets, making level decisions quantifiable. The resulting rule links higher levels to category features and instance-of, while treating its feature influence as partial inheritance.
Problem
MLM links levels through non-transitive instance-of membership, yet its approaches support deep characterization that gives category classes superclass-like influence.
Source: §1 · pp. 1–2, §3.1 · p. 7Method
CatMLM gives category classes dual powertype and superclass facets by separating non-inheriting category features from object features that participate in deep characterization.
Source: §1 · p. 2Results
Category features justify a new abstraction level and instance-of relation, providing a quantifiable decision rule for leveling.
Source: §1 · p. 2Takeaways & Limitations
CatMLM extends class-hierarchy abstraction through category classes’ superclass facet and frames instance-of as partial inheritance of object features.
Source: §1 · p. 2The model assumes category-class features divide into object features that participate in deep characterization and category features that do not.
Source: §1 · p. 2
Abstract
from arXiv · showhide
MultiLevel software Modeling (MLM) suggests that conceptual modeling in broad subject domains might require abstraction of multiple classification levels. The MLM approach relies on philosophical arguments, claiming that faithful modeling of real-world domains involves repeated type classification as in ontologies of natural kinds. MLM leveled architecture is interwoven and defined by instance-of interlevel relationships between clabject classes in lower levels to classes termed category classes, in upper levels. The instance-of relation denotes membership of clabjects as type objects in their (powertypes) category classes, and is not transitive. All MLM approaches support forms of deep characterization, i.e., category classes can influence classes in lower levels. Deep characterization is an essential feature of superclasses and contradicts the non-transitive membership meaning of instance-of. In this paper, we introduce the Category-Based MLM (CatMLM) model, in which category classes have dual superclass and powertype facets, based on the distinction between category features that do not participate in deep characterization, and object features that do. This distinction clarifies the role of levels, provides a clear quantifiable criterion for leveling, and yields a decision rule between the subclass and instance-of relations. The contribution of this paper is to introduce a well-defined MLM model that (1) is based on simple, quantifiable level decisions; (2) clarifies how leveling emerges from domain needs; and (3) analyzes gains and losses of MLM vs. plain OO modeling.
1 Introduction
MLM uses multiple classification levels linked by instance-of relationships, but the non-transitive membership relation sits uneasily with the deep characterization that MLM approaches support. CatMLM distinguishes category and object features to give category classes both powertype and superclass facets and to make leveling decisions explicit.
1 Introduction
- MLM proposes multiple classification levels for broad domains, drawing on the claim that real-world modeling involves repeated type classification.Instance-of pairs a lower-level clabject, or class that is also an object, with an upper-level category class.Source: §1 · p. 1
- CatMLM assigns category classes dual powertype and superclass facets, distinguishing category features from object features that participate in deep characterization.Object features apply to data objects of offspring classes across levels; category features characterize instances of the category class alone.Source: §1 · p. 2
- Deep characterization lets category classes influence offspring classes, a superclass-like effect not inherent in instance-of membership.The paper describes this influence across chains of instance-of and subclass relationships.Source: §1 · p. 2
- Category features mark a new level of abstraction and provide a decision rule: use instance-of when they justify a higher level, otherwise use subclassing.The proposed rule connects level introduction to the category features of category classes.Source: §1 · p. 2
- CatMLM treats instance-of as partial inheritance of object features, unlike subclassing’s full inheritance and associations’ lack of inheritance.The model extends class-hierarchy abstraction through category classes’ superclass facet.Source: §1 · p. 2
- CatMLM is presented as a formal model with quantifiable level decisions that clarifies how domain needs shape levels and extends class-hierarchy abstraction.It simplifies MedMLM by replacing complex interlevel mediators with the category/object feature distinction.Source: §1 · p. 2
2 Category-based MLM
CatMLM treats category classes as both powertypes and superclasses, separating features that introduce abstraction from those that characterize offspring. This distinction guides level structure, inheritance, and the model’s object semantics.
2 Category-based MLM
- Category classes have a superclass facet for object features that characterize offspring and a powertype facet for category features that are not inherited.Object features follow OO inheritance, whereas category features apply only to objects of the category class.Source: §2 · pp. 3–4
- Instance-of relates classes in different levels, and CatMLM uses category features to justify introducing a category class and a new abstraction level.A level is a plain class model; instance-of relationships mark distinctions between levels.Source: §2 · pp. 3–4
- CatMLM distinguishes category, object, and interlevel associations and constraints by which objects they apply to and whether they are inherited.Object features are inherited by clabjects, while category features are not; object-typed operations and constraints can use the category class’s object type.Source: §2 · pp. 4, 6
- Classes on internal levels, including category classes, have their own data objects, and clabject objects link to objects of each of their category classes.These links let clabject objects retrieve their specific category information.Source: §2 · p. 5
- CatMLM’s intended instances combine legal object populations across levels with interlevel links, instance-of links, and constraints.Each clabject object must link to an object of each of its category classes.Source: §2 · p. 5
- The model treats instance-of as partial inheritance of object features, unlike subclassing’s full inheritance and associations’ lack of inheritance.Object-typed features inherited via instance-of satisfy the Liskov substitution principle; category-typed features do not.Source: §2 · p. 6
3 Related Works
Related MLM approaches vary in classification dimensions and implementation strategies. The section reviews deep-characterization mechanisms and contrasts CatMLM’s feature inheritance with MedMLM’s.
3 Related Works
- Most MLM approaches use one classification dimension; multidimensional approaches add orthogonal hierarchies, while tools range from native MLM environments to extensions of two-level modeling tools.CatMLM is planned as an extension of MLM-USE.Source: §3 · p. 7
3.1 Deep Characterization in MLM
- Deep characterization lets category classes influence clabject classes, selectively applying feature inheritance associated with subclass-superclass relations.Potency marking is a dominant approach, assigning elements an influence distance across levels.Source: §3.1 · p. 7
- Deep-characterization alternatives include potency variants, deep constraints, and default or selected feature inheritance.Some systems use interlevel keywords for constraints; others support refining or leaving attributes partially uninstantiated until lower levels.Source: §3.1 · p. 7
- MedMLM inherits all category-class features by default unless prevented, whereas CatMLM inherits object features but not category features.Both models support interlevel associations and constraints; CatMLM can also derive influence operations from its inheritance mechanisms.Source: §3.1 · p. 7
4 Conclusions and Future Research
The conclusion presents CatMLM’s dual category-class facets and its category/object feature distinction as the basis for type refining and level structuring. Future work plans tool support and a formal typing theory for instance-of.
4 Conclusions and Future Research
- CatMLM uses category/object feature distinctions for type refining, level structuring, model development, and extending subclassing’s computational abstraction.The model is based on category classes’ dual superclass and powertype facets.Source: §4 · p. 7
- Future work plans to extend an existing MLM tool for CatMLM and develop a formal instance-of typing theory with Category_object and Category_category distinctions.Source: §4 · p. 7