A diagrammatic reasoning system for the description logic ALC

https://doi.org/10.1016/j.jvlc.2007.12.003Get rights and content

Abstract

Diagrammatic reasoning is a tradition of visual logic that allows sentences that are equivalent to first order logic to be written in a visual or structural form: usually for improved usability. A calculus for the diagram can then be defined that allows well-formed formulas to be derived. This calculus is intended in the analog of logical inference.

Description logics (DLs) have become a popular knowledge representation and processing language. DLs correspond to decidable fragments of first order logic; their notation is in the style of symbolic, variable-free formulas. Moreover, DLs are equipped with table au theorem provers that are proven to be sound and complete.

Although DLs have roots in diagrammatic languages (such as semantic networks), they are elaborated in a purely symbolic manner. This paper discusses how DLs can be equivalently represented in terms of a diagrammatic reasoning system.

First, existing diagrammatic reasoning systems, namely spider- and constraint diagrams, as well as existential and conceptual graphs, are investigated to determine if they are compatible with DLs. It turns out that Peirce's existential graphs are better suited for this purpose than the alternatives we examine.

The paper then redevelops the DL ALC, which is the smallest propositional DL, by means of labeled trees, and provides a diagrammatic representation for these trees in the style of Peircean graphs. We provide a calculus based on C.S. Peirce's calculus for existential graphs and prove the soundness and completeness of the calculus. The calculus acts on labeled trees, but can be best understood as a diagrammatic calculus whose rules modify the Peircean-style representation of ALC.

Introduction

Description logics (DLs) are a common family of knowledge representation languages tailored to express knowledge about concepts and concept hierarchies. They include sound and complete decision procedures for reasoning about such knowledge. One of the main applications of DLs is their use as the basis for an ontology language, especially popular for the Semantic Web. In particular, the Ontology Web Language (OWL)—a W3C recommendation for the knowledge language of the Semantic Web—is based on a specific and expressive DL termed SHOIN(D).1

The basic building blocks of DLs are concepts (unary predicates), roles (binary relations) and sometimes individuals, which can be composed by language constructs such as intersection, union, value or number restrictions to build more complex well-formed formulas that themselves represent complex concepts and roles. For example, when man, female, male, rich, happy are predefined concepts and if haschild is a predefined role, thenMANHASCHILD.FEMALEHASCHILD.MALEHASCHILD.(RICHHAPPY)describes the concepts of men who have both male and female children, and where all the children are rich or happy. Let us call a concept defined in this way as happyman.

The formal notation of DLs has the flavour of a variable-free first order predicate logic (FOL). In fact, DLs correspond to (decidable) fragments of FOL, and like FOL, DLs have a well-defined, formal syntax, a semantics in the form of Tarski-style models, and a sound and complete calculi (based on Table aux-algorithms). It is often emphasised that DLs offer, in contrast to other knowledge representation languages, sound, complete and (empirically) tractable reasoning services. A comprehensive overview on DLs is given in the Description Logic Handbook [2].

The notation of DLs is in the style of the usual linear and symbolic2 notations of FOL. The fact that the notation of DLs is variable-free makes them easier to comprehend than the common FOL formulas which include free variables. Nonetheless, for untrained users, the symbolic notation of DLs can be hard to learn and comprehend.

A main alternative to the symbolic notation is the development of a diagrammatic representation of DLs. It is well accepted that diagrams are in many cases easier to comprehend than symbolic notations (see for example [3], [4], [5], [6]), and in particular it has been argued that they are useful for knowledge representation systems [7], [8]. This has been acknowledged by the DL community and is a common view among the broader knowledge representation community [9]. In [10], the introduction to the Description Logic Handbook, Nardi and Brachman write that besides the possibility of “providing a syntax that resembles more closely natural language”, a “major alternative for increasing the usability of DL as a modeling language” is to “implement interfaces where the user can specify the representation structures through graphical operations.”

A first attempt at a diagrammatic representation for DL is can be found in [7], where Gaines elaborates a graph-based representation for the textual DL CLASSIC, part of the kl-one-framework. More recently, the focus has shifted from the development of proprietary diagrammatic representations to representations within the framework of UML (unified modeling language). In 2003, the Object Management Group requested a metamodel for the purpose of defining ontologies. Following this proposal, Brockmans et al. [11] provide a UML-based, diagrammatic representation for the OWL DL. In these approaches, the focus is on a graphical representation of DL, however, as emphasized in many works on DL (see for example [2]), reasoning is seen as a distinguishing feature of DL and such reasoning is not supported diagrammatically by that treatment. Correspondences between graphical representation of the DL and the DL reasoning system are therefore important but remain largely unelaborated to date. Similar arguments hold for other popular diagrammatic languages like UML and ORM3 the difference being that, that unlike DLs, these diagrammatic modeling languages provide no extensional, mathematical semantics, nor any automated reasoning facilities.

On the other hand, there are some candidate diagrammatic reasoning systems that have the expressiveness of fragments of FOL, or even full FOL. In this paper, we will evaluate two families of contemporary mathematical diagrammatic reasoning systems, which have the following two (historical) origins:

  • (i)

    The system of Euler circles and Venn-diagrams, the latter enriched by C.S. Peirce to Venn-Peirce-diagrams (see [12], [13]).

  • (ii)

    The system of Peirce's Existential Graphs (EG).4

The first system is the background to the contemporary development of spider diagram (SD) and constraint diagrams (CD), the latter is background to a contemporary interpretation as conceptual graphs (CG) [14]. Why is it worth considering these diagrammatic reasoning systems as a starting point for a diagrammatic version of DL? Firstly, for all these systems mathematical elaborations in the general style of mathematical logic exists. These include:
  • A well-defined syntax: usually, the syntax is defined at an abstract level (for example in terms of graph theory), such that the well-formed formulas have diagrammatic representations. Moreover, the syntax is—like the syntax for DL—variable-free.

  • Extensional—Tarski-style semantics—and/or translations to formulas of FOL.

  • Sound and complete calculi: usually the rules are defined on the abstract syntax but mostly they can be understood as manipulations of the represented diagrams.

Therefore, in contrast to data modeling languages like UML and ORM, we have a well-defined syntax and semantics for the candidate diagrammatic representations. Moreover, if we adopt one of these systems, we can adopt (partly or completely) the existing calculi as a diagrammatic reasoning service for DL. Finally, over and above the general arguments for diagrammatic systems, some of the systems investigated in this paper—constraint and SD and CGs—have had their diagrammatic benefits investigated in user-evaluations [15], [16], [17], [18].

As a specific DL is the corresponding logic behind the OWL, the results of this paper benefit not only DL, but the Semantic Web more generally. Developing a Semantic Web language as a mathematical diagrammatic reasoning system has already been carried out for a much simpler language: namely RDF. For RDF, mathematical elaborations based on graph theory, including a Tarski-style semantics and a sound and complete calculus based on “projections” (see [19], [20]) or via diagrammatic rules (see [21]) have been elaborated.

In the next three 2 Introduction into DL, 3 Spider and constraint diagrams, 4 Existential and conceptual graphs, the basic notions of DL, SD and CD, and EGs and CGs, are introduced. In the subsequent 5 SD or CD for DL, 6 Conceptual or EGs for DL, we investigate whether SD and CD and EGs and CGs, respectively, are suited as a starting point for developing DL as a diagrammatic reasoning system. It will turn out that there are prospects to SD and CD for this purpose. However, CGs and, as we will show, EG are a much better match. In the three sections that follow, ALC is developed as a diagrammatic reasoning system based on Peirce's EGs. In Section 7, a formalization of ALC by means of labeled trees is given. Section 8 provides the Peirce style representation of such trees. In Section 9, a calculus for ALC-trees, based on Peirce's calculus for EGs, is given, and soundness and completeness proven. Finally, in Section 7, further research is discussed. As we use several abbreviations in this paper, these are presented in Appendix.

Section snippets

Introduction into DL

The vocabulary of any DL consists of concepts, which denote sets of individuals, and roles, which denote binary relationships between individuals. The starting points are atomic concepts and rolesconcept and role names—from which more complex concepts and roles are built with constructs such as intersection, union, value or number restrictions, and so on. Moreover, we usually consider concept vocabularies where we have a universal concept . The tables of Fig. 1, Fig. 2 provide an overview of

Spider and constraint diagrams

SDs and CDs are a diagrammatic reasoning system based on Euler circles and Venn–Peirce diagrams. They are inspired by a fragment of UML which is used for software specifications as an alternative to UML's textual object constraint language (OCL). They are elaborated as abstract mathematical structures, including an extensional semantics and inference rules. In this section, SDs and CDs are briefly and informally introduced. Particularly, only the diagrams of SDs and CDs are provided, not the

Existential and conceptual graphs

In this section, we introduce Peirce's EGs, Sowa's CGs, particularly its most common fragment of simple conceptual graphs (SCGs) and the system of concept graphs with cuts (CGwCs) introduced by one of the authors. Similar to SDs and CDs, there exist mathematically precise elaborations for each of these systems. Again, like SDs and CDs, the diagrammatic benefits of EGs or CGs (used in teaching) has been investigated by Schäfe [16], [84] and also by Pollant [15]. Furthermore, a theorem prover for

SD or CD for DL

At a first glance, there are some striking similarities between DL and the system of SDs and CDs. Most importantly, both correspond to decidable fragments of FOL where only unary predicates and binary relations are used. Both systems have sound and complete calculi which are implemented on tableaux-based algorithms. It is therefore worth investigating whether SD and CD can be used as a diagrammatic representation for DL.

In our scrutiny of SDs and CDs, we first restrict ourselves to a discussion

Conceptual or EGs for DL

Both CGs and DLs have some common background: they are both developed and intended as knowledge representation systems to include reasoning facilities, they were developed at nearly the same time and they have an important common ancestor, namely the semantic networks of AI (and Minski's frame systems). In fact, in 1979, the well-known system kl-one was developed on the basis of informal use of semantic networks in AI (see [60], [61]). kl-one was itself a diagrammatic formalism and in 1987 a

A tree-based formalization of the DL ALC

In this section, the syntax and semantics for the DL ALC is provided. First of all, as mentioned earlier in Section 2, the DL ALC has conjunction, negation and value restriction as constructors. As the RGs of Peirce have conjunction, negation and existential quantification as constituents, we can consider ALC made up from conjunction, negation and existential restriction (instead of value restriction).

Let RGALC denote the class of RGs that will be used to formalize ALC. The graphs of RGALC will

A diagrammatic representation for ALCTree

In Section 4.1, we briefly discussed that C.S. Peirce distinguished between graphs and graph replicas. As it is more generally argued in [42], [43], any formalization of logic by means of diagrams has to distinguish these two levels, usually called types or abstract syntax (which correspond to Peirce's graphs), and tokens or concrete syntax (corresponding to (Peirce's replicas). The ALC-trees, as they have been defined in the last section, are obviously the abstract syntax. In this section we

The calculus for ALCTree

C.S. Peirce provided a set of five rules for the system of EGs termed erasure, insertion, iteration, deiteration, double cut. It is possible to elaborate Peirce's graphical calculus, including a proof of its soundness and completeness, in a mathematical precise manner, and to extend the calculus for the system of RGs, but this goes beyond the scope of this paper.

RGALC is a fragment of the full system of RGs. As one would expect, the rules for RGs are also sound rules for RGALC. However it is not

Conclusion and further research

This paper has discussed different diagrammatic reasoning systems—spider and constraint diagrams on the one hand, existential and conceptual graphs on the other—as candidates for a diagrammatic version of DL. It has been argued that spider and constraint diagrams are not as well suited to a diagrammatic version of DLs as conceptual graphs and existential graphs. More specifically, a version of conceptual graphs including negation and free variables, and relation graphs—existential graphs with

Acknowledgements

We like to thank the reviewers for their helpful remarks and also Andrew Fish for his valuable comments on using CDs for ALC.

References (86)

  • S.-J. Shin

    The Iconic Logic of Peirce's Graphs

    (2002)
  • A. Shimojima, On the Efficacy of Representation, Ph.D. thesis, The Department of Philosophy, Indiana University, 1996....
  • S. Berger, Studies on the uses and usefulness of diagrams, Master's thesis, Institute for Logic, Language and...
  • B.R. Gaines, An interactive visual language for term subsumption languages, In: IJCAI, 1991, pp....
  • R. Kremer, Visual languages for knowledge representation, in: Proceedings of 11th Workshop on Knowledge Acquisition,...
  • D. Nardi, R.J. Brachman, An introduction to description logics, in: Baader et al. [2], pp....
  • S. Brockmans, R. Volz, A. Eberhart, P. Löffler, Visual modeling of owl dl ontologies using uml, in: S.A. McIlraith, D....
  • S.-J. Shin

    The Logical Status of Diagrams

    (1994)
  • E.M. Hammer

    Logic and Visual Information

    (1995)
  • J.F. Sowa

    Conceptual structures: information processing in mind and machine

    (1984)
  • A. Vile, S. Polovina, Possibilites in peirce's existential graphs for logic education, available from the authors,...
  • H. Schärfe, U. Petersen, P. Øhrstrøm, On teaching conceptual graphs, in: Priss et al. [84], pp....
  • A. Fish, J. Masthoff, An experimental study into the default reading of constraint diagrams, In: VL/HCC, IEEE Computer...
  • A. Fish, J. Masthoff, Do monkeys like elephants or do elephants watch monkeys? An empirical study into the default...
  • J.-F. Baget

    Homomorphismes d’hypergraphes pour la subsumption en rdf/rdfs

    Langages et modèles à objets 2004 (actes 10e conférence), RSTI - L’objet (numéro spécial)

    (2004)
  • J.-F. Baget, Rdf entailment as a graph homomorphism, in: Yolanda Gil, Enrico Motta, V. Richard Benjamins, Mark A. Musen...
  • F. Dau, Rdf as graph-based, diagrammatic logic: Syntax, semantics, calculus, normalforms, in: F. Esposito, Z.W. Ras, D....
  • G. Stapleton, Reasoning with Constraint Diagrams, Ph.D. Thesis, Visual Modelling Group, Department of Mathematical...
  • G. Stapleton et al.

    Spider diagrams

    LMS Journal of Computation and Mathematics

    (2005)
  • J. Gil, J. Howse, S. Kent, Formalizing spider diagrams, in: IEEE Symposium on Visual Languages, 1999, pp....
  • VMG. The visual modeling group homepage, university of Brighton...
  • J. Flower, J. Masthoff, G. Stapleton, Generating proofs with spider diagrams using heuristics, in: International...
  • J. Flower, J. Masthoff, G. Stapleton, Generating readable proofs: a heuristic approach to theorem proving with spider...
  • O. Patrascoiu, S. Thompson, P. Rodgers, Tableaux for diagrammatic reasoning, in: P. Cox, T. Smedley (Eds.), Proceedings...
  • S. Kent, Constraint diagrams: Visualizing assertions in object-oriented models, in: OOPSLA, ACM Press, 1997, pp....
  • J. Stewart, Theorem proving using existential graphs, Master's Thesis, University of California, Santa Cruz, 1996....
  • W. Hartshorne, Burks, editors, Collected Papers of Charles Sanders Peirce, Harvard University Press, Cambridge, MA,...
  • D.D. Roberts

    The Existential Graphs of Charles S. Peirce

    (1973)
  • J.J. Zeman, The Graphical Logic of C.S. Peirce, Ph.D. Thesis, University of Chicago, 1964, Available at:...
  • C.S. Peirce, J.F. Sowa, Existential Graphs: MS 514 by Charles Sanders Peirce with commentary by John Sowa, 1908, 2000,...
  • J.F. Sowa, Logic: Graphical and algebraic, manuscript, Croton-on-Hudson,...
  • F. Dau, Mathematical logic with diagrams, based on the existential graphs of peirce, Habilitation thesis. To be...
  • C.S. Peirce, MS 478: Existential Graphs, Harvard University Press, 1931–1935. Partly published in of [34] (4.394-417)....
  • Cited by (23)

    • On a graph calculus for modalities

      2017, Theoretical Computer Science
      Citation Excerpt :

      In terms of expressive power, nominals and our nodes are closely related; an important difference resides in the visual appeal provided by our 2-dimensional notation (cf. p. 82). Other graphical approaches to logic, e.g. [9], are based on Peirce's work [17,18]. Though sharing the visual motivation, these approaches differs from ours in some aspects: while Peirce's approach may be termed geometrical, ours is more algebraic; also Peirce's rules give entailment, whereas our rules involve equivalence (cf. Remark 5.1, so they can be used in any context).

    • Verifying a medical protocol with temporal graphs: The case of a nosocomial disease

      2014, Journal of Critical Care
      Citation Excerpt :

      As a result, the possibility to interact and exchange knowledge modeled with internal and external staff is available to different stakeholders in the modeling system under this consideration. In addition to the visual reasoning facilities, syntactic opportunities for graphs include the intuitive ideas of graphical arguments that have no equivalent in DL [17]. However, the expressiveness of conceptual graphs may not be satisfactory for modeling the temporal knowledge of a given domain.

    • Conceptual graph-based knowledge representation for supporting reasoning in African traditional medicine

      2013, Engineering Applications of Artificial Intelligence
      Citation Excerpt :

      As well as similarities between DL and CGs, there are specific interesting features: cycles, n-ary relations, and type-hierarchy for CGs; the style of symbolic, variable-free formulas, variety of constructors with different levels of expressiveness for DLs. Besides visual reasoning facilities, the syntactical possibilities of the graphs, including identity, graphs that contain circles, graphs that are not connected, etc., allow graphs to be constructed that do not have counterparts in DL (Dau and Eklund, 2008). However, the conceptual graphs’ expressiveness might not be satisfactory for temporal modeling of a given domain knowledge.

    • Visualizing ALC using concept diagrams

      2018, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    View all citing articles on Scopus
    View full text