A Survey on how Description Logic Ontologies Benefit from Formal Concept Analysis

Although the notion of a concept as a collection of objects sharing certain properties, and the notion of a conceptual hierarchy are fundamental to both Formal Concept Analysis and Description Logics, the ways concepts are described and obtained differ significantly between these two research areas. Despite these differences, there have been several attempts to bridge the gap between these two formalisms, and attempts to apply methods from one field in the other. The present work aims to give an overview on the research done in combining Description Logics and Formal Concept Analysis.

💡 Research Summary

The paper provides a comprehensive survey of research that bridges Description Logics (DLs) and Formal Concept Analysis (FCA), two formalisms that both deal with the notion of a “concept” and a hierarchical organization of concepts, yet approach them from fundamentally different perspectives. DLs are a family of knowledge‑representation languages used to model the terminological (TBox) and assertional (ABox) components of an ontology. Concepts are built from atomic concepts (unary predicates) and atomic roles (binary predicates) using constructors such as conjunction (⊓), disjunction (⊔), negation (¬), universal restriction (∀), and existential restriction (∃). The semantics is given by interpretations that map each concept to a subset of a domain and each role to a binary relation. Standard DL reasoning services include subsumption checking, instance checking, and consistency verification, typically realized by tableau‑based or hypertableau reasoners (e.g., FaCT++, Pellet, Hermit).

In contrast, FCA starts from a formal context – a binary incidence matrix between objects and attributes – and defines a formal concept as a pair (extent, intent) where the intent is the set of attributes common to all objects in the extent, and the extent is the set of objects sharing all attributes in the intent. The set of all formal concepts forms a complete lattice (the concept lattice), which provides a visual and algorithmic tool for exploring the structure of the data. FCA’s intensional descriptions are limited to conjunctions of atomic attributes; there is no built‑in notion of roles, quantifiers, or complex constructors.

The survey identifies two major research directions that aim to combine the strengths of both formalisms.

1. Enriching FCA with DL constructors – Several works have extended the expressive power of FCA by borrowing DL operators. Prediger and Stumme introduced “logical scaling”, where DL expressions are used to define constraints on attribute combinations; a DL reasoner acts as an expert during attribute exploration, providing counterexamples when an implication fails. Prediger’s “terminological attribute logic” (t‑AL) essentially embeds the DL ALC language into FCA, adding existential/universal quantifiers and negation to attributes. Relational Concept Analysis (RCA) by Rouane et al. builds a family of related formal contexts and connects them via binary relations, then maps the resulting concepts into a sub‑language of ALE called FL‑E (allowing ⊓, ∀, ∃, ⊤, ⊥). After the mapping, DL reasoning is employed to check consistency and classify the concepts. These approaches demonstrate that FCA can be made more expressive, allowing it to model relational data and richer ontological constraints.

2. Applying FCA methods to DL problems – The second line of work uses FCA’s algorithmic machinery to solve DL‑centric tasks. Baader showed that the subsumption hierarchy of all possible conjunctions of a set of DL concepts can be obtained by constructing a formal context whose attributes are the defined concepts and whose objects are counter‑examples (interpretations). The resulting concept lattice is isomorphic to the desired subsumption lattice, enabling the discovery of hidden inclusion relationships (e.g., a conjunction of two concepts may be subsumed by a third even when the individual concepts are incomparable). Stumme extended this to include disjunctions, using distributive concept exploration to compute the full lattice of conjunctions and disjunctions. Baader and Molitor applied FCA to compute the hierarchy of least common subsumers (LCS) for a set of examples, supporting bottom‑up ontology engineering where examples guide the synthesis of a concept description. Rudolph’s “relational exploration” integrates DLs (typically EL or FL‑E) into FCA by defining DL‑based attributes and using FCA’s interactive knowledge acquisition to refine DL knowledge bases; binary power‑context families provide a semantic foundation for representing relational information.

The paper emphasizes that DLs bring a rich, decidable logical language and powerful automated reasoning, while FCA offers a data‑driven, exploratory view of the entire concept space via lattices. By combining them, one can use DL reasoners to validate and enrich FCA‑derived structures, and conversely employ FCA’s attribute exploration to systematically uncover subsumption relations, LCS hierarchies, or to guide the construction of ontological axioms. The surveyed works span applications in medical informatics, software engineering, configuration management, and the Semantic Web, illustrating that the DL–FCA synergy can improve ontology quality, support interactive knowledge acquisition, and enable more scalable reasoning over large, relational datasets. The survey concludes that continued cross‑fertilization between these fields is promising for both theoretical advances and practical ontology engineering.